Ecohydrological modelling with EcH2O‐iso to quantify forest and grassland effects on water partitioning and flux ages

We used the new process‐based, tracer‐aided ecohydrological model EcH2O‐iso to assess the effects of vegetation cover on water balance partitioning and associated flux ages under temperate deciduous beech forest (F) and grassland (G) at an intensively monitored site in Northern Germany. Unique, multicriteria calibration, based on measured components of energy balance, hydrological function and biomass accumulation, resulted in good simulations reproducing measured soil surface temperatures, soil water content, transpiration, and biomass production. Model results showed the forest “used” more water than the grassland; of 620 mm average annual precipitation, losses were higher through interception (29% under F, 16% for G) and combined soil evaporation and transpiration (59% F, 47% G). Consequently, groundwater (GW) recharge was enhanced under grassland at 37% (~225 mm) of precipitation compared with 12% (~73 mm) for forest. The model tracked the ages of water in different storage compartments and associated fluxes. In shallow soil horizons, the average ages of soil water fluxes and evaporation were similar in both plots (~1.5 months), though transpiration and GW recharge were older under forest (~6 months compared with ~3 months for transpiration, and ~12 months compared with ~10 months for GW). Flux tracking using measured chloride data as a conservative tracer provided independent support for the modelling results, though highlighted effects of uncertainties in forest partitioning of evaporation and transpiration. By tracking storage—flux—age interactions under different land covers, EcH2O‐iso could quantify the effects of vegetation on water partitioning and age distributions. Given the likelihood of drier, warmer summers, such models can help assess the implications of land use for water resource availability to inform debates over building landscape resilience to climate change. Better conceptualization of soil water mixing processes and improved calibration data on leaf area index and root distribution appear obvious respective modelling and data needs for improved simulations.


| INTRODUCTION
Vegetation exerts a strong control on land-surface water and energy partitioning, and the resulting ecohydrological fluxes of "green water" as evaporation and transpiration (Baldocchi, Xu, & Kiang, 2004;Llorens & Domingo, 2007;Villegas et al., 2015;Wang, Good, & Caylor, 2014), and the residual "blue water" fluxes to groundwater recharge and run-off (Jencso & McGlynn, 2011;Williams et al., 2012). Thus, it is well established that forests generally "use" more water than grass or short crops due to higher evapotranspiration (Fatichi, Pappas, & Ivanov, 2016). However, there are strong codependencies between plant growth, seasonal phenological change and life cycles, and the physical processes that drive water movement that we poorly understand (Jolly & Running, 2004;Porporato, Laio, Ridolfi, & Rodriguez-Iturbe, 2001;Wang, Tetzlaff, Dick, & Soulsby, 2017). Elucidating these interactions is an essential prerequisite for modelling the hydrological effects of land cover change and their ecological implications and for mitigating the negative impacts that it may have, including effects on climate at local, regional, and global scales. Thus, although it is well known that conversion of natural forest to agriculture generally reapportions "green" and "blue" water fluxes, in the direction of increasing run-off, intensifying downstream flood risk and enhancing baseflows; the effects are usually site specific depending on hydroclimate and biogeography (Amogu et al., 2015;Archer, 2007;García-Ruiz et al., 2008). In addition, climate change predictions generally indicate higher future atmospheric water demands in temperate regions, as well as seasonal redistribution of rainfall with potentially drier summers (Trenberth et al., 2007). Within this context of growing climatic stress, the hydrological implications of land cover change on water partitioning and water availability are of increasing concern (Frei, Schöll, Fukutome, Schmidli, & Vidale, 2006;Nikulin, Kjellström, Hansson, Strandberg, & Ullerstig, 2011). Linked to this, in many areas, the natural response of vegetation communities to climate change is already being observed (Menzel et al., 2006), and this may also have important hydrological impacts (Tetzlaff et al., 2013).
Yet despite the long history of research into land cover change effects on hydrology, many details of the role of vegetation in regulating water partitioning are difficult to quantify (Zhang, Yang, Yang, & Jayawardena, 2016). For example, the impacts of vegetation dynamics on water fluxes-in terms of directional long-term growth and seasonal phenology-are rarely well constrained (Huisman et al., 2009).
Along with advances in ecohydrological modelling, experimental studies using water isotopes and other conservative tracers have advanced our understanding of how water flows in catchments, and helped improve how hydrological models represent the celerity of hydrological fluxes, as well as the velocity of water particles and the mixing relationships within soils (McGuire & McDonnell, 2006;Birkel, Soulsby, & Teztlaff, 2011;Peters, Burns, & Aulenbach, 2014;Klaus, Chun, McGuire, & McDonnell, 2015;Benettin, Kirchner, Rinaldo, & Botter, 2015;Sprenger, Tetzlaff, Buttle, Carey, et al., 2018). Such tracer-aided models can also track the age distributions of water in different catchment storage compartments and "green" and "blue" water fluxes (Soulsby et al., 2015;van Huijgevoort, Tetzlaff, Sutanudjaja, & Soulsby, 2016;Remondi, Kirchner, Burlando, & Fatichi, 2018). Recent work has shown that when integrated with explicit representation of vegetation dynamics, these tracer-aided modelling concepts are better positioned to assess the interactions of plants and water partitioning in response to hydroclimatic variability because they can help determine the pools of water that plant use and how they affect water mixing, hydrologic connectivity, and the establishment of flow paths (Kuppel, Tetzlaff, Maneta, & Soulsby, 2018b). In conjunction with tracer data, such models can provide insight into the fate of soil water and the processes that determine the ecohydrological separation of "green" fluxes that sustain biomass and "blue" water fluxes that sustain groundwater recharge and stream flow generation (Evaristo, Jasechko, & McDonnell, 2015;McDonnell, 2014;. In this study, we apply the process-based, tracer-aided ecohydrological model, EcH 2 O-iso (Kuppel et al., 2018b), to quantify the contrasting effects of seminatural forest and grassland on water partitioning and flux ages at an intensively studied site in Northern Germany. The data-rich nature of the site provided measurements of energy balance components (through a fully automated weather station), hydrological processes (precipitation, throughfall, transpiration, soil moisture, and groundwater levels) and biomass production (litterfall and forest growth metrics). The study site at Stechlin is located in the German state of Brandenburg, which is droughtsensitive, and is an extensively forested area where there is concern over land use effects on groundwater recharge; in particular, the role of forest management for timber production in reducing recharge. This is compounded by climate change predictions, which forecast warmer and drier summers (Dorau, Gelhausen, Esplör, & Mansfeldt, 2015;Lischeid & Nathkin, 2011;Riediger, Breckling, Svoboda, & Schröder, 2016). Thus, we aim to show how multiproxy data can be used to improve the robustness of quantifying land use effects on the water balance using an ecohydrological model. This is done with the anticipation that such advances will help inform land use strategies designed to build landscape resilience to climate change and protect water resource needs. The specific objectives are the following: 1. To conduct a multicriteria calibration of EcH 2 O-iso for concurrent energy balance, water balance, and biomass production simulations.
2. To quantify, within an uncertainty framework, the role of forest and grassland vegetation on the local water balance in terms of water partitioning and ages of different vertical fluxes.
3. To use tracers to test the internal consistency of the model and implications for interpreting the resulting age distributions of different fluxes.
4. To assess the implications of future land management, in the context of climate change, for water partitioning and water availability.
The context of the modelling was also to use EcH 2 O-iso in a learning framework to understand how to both improve the model and prioritize data collection for future studies.
2 | STUDY SITE AND DATA

| Site description
Two long-term study plots are located in the catchment of the extensively monitored Lake Stechlin (Bergström et al., 2003;Casper, 2012;Dieffenbach-Fries, Hofmann, & Schleyer, 2003); a groundwater-  (Thornthwaite, 1948). The dominant land cover in the area is >100-year-old seminatural mixed deciduous/conifer forest, which is managed for conservation purposes. In the forest study plot, FIGURE 1 Study site location of the two monitoring plots in Northern Germany and aerial view of edge of Lake Stechlin (a), the topography (b), view of the forest (c) and grassland plots (d), and schematic profile section of ground surface elevation, soil profile characteristics (the soil horizons are defined according to the USDA taxonomy: a-c are the surface, illuvial, and little-altered substrate horizons, respectively; the suffixes h, t, and v refer to organic rich, clay translocation, and iron deposition, respectively), water table depth and instrumentation depths (e). Photos (c) and (d) from Brüning, Graf, & Nützmann, 2003 FIGURE 2 (a) Main hydroclimatic data plotted at monthly time steps: Net solar radiation (yellow), monthly average of daily minimum, maximum and average temperature (red dashed and solid lines), and monthly average of the air water deficit (blue cyan line); (b) precipitation time series: number of prior days with less than 10 mm precipitation (black line) and monthly and yearly accumulation (blue hyetograms and grey polygons); daily soil water content (c) in the forest and (d) grass plots this is mostly composed of deciduous beech (Fagus sylvatica, >80%) trees and also some Scots pine (Pinus sylvestris; Table 1; Schulte-Bisping, Beese, & Dieffenbach-Fries, 2012). The second plot is perennial semidry grassland dominated by a dense sward of Calamagrostis epigejos, Festuca ovina, and Koeleria glauca, which is cut (similar to a cut or grazed meadow) to facilitate access to the instrumentation ( Figure 1). This is typical grassland vegetation after forest removal and a grazing management regime.
The plots are in a lowland area (~65-m elevation) with a gentle slope (median = 4.5%). The lake and low-lying topography were formed at the end of the Weichselian glacial period, with extensive drift deposits covering the solid geology (Merz & Pekdeger, 2011).

| Available data
The plots form part of long-term environmental monitoring at Stechlin (Bergström et al., 2003;Casper, 2012;Dieffenbach-Fries et al., 2003;Pöschke, Nützmann, Engesgaard, & Lewandowski, 2018) and are linked to a network of pan-European sites for assessing the ecosystems effects of acid deposition (Tørseth et al., 2012;Fagerli & Aas, 2008). Consequently, a wide range of data is available for ecohydrological modelling (Figure 3). Long-term hourly climate observations (precipitation, temperature, incoming short-wave radiation, outgoing long-wave radiation, relative humidity, wind speed, etc.) have been measured adjacent to the grassland plot since 1950. Soil properties have been characterized, and soil moisture has been monitored since 2000 (Nützmann, Holzbecher, & Pekdeger, 2003). TDR probes were installed in 2004 at four depths (three replicates) at both plots: 30, 50, 70, and 250 cm at the grassland and 30, 50, 120, and 350 cm at the forest. In the latter part of the study, soil temperatures (from the TDR probes) have also been monitored. Technical problems prevented data collection at the forest plot for a just over a year between 2010 and 2011.
At the forest plot, biomass measurements related to seasonal and long-term tree growth dynamics are available. Litterfall was measured between 2004 and 2009, allowing an annual assessment of leaf, needle, and fruit production. Stem diameters of 12 representative trees were measured during the same period using dendrometer bands.
Using the tree volume equations of Bergel (1973), this provides a first approximation of the above-ground biomass accumulation. Unfortunately, equivalent biomass data were not available for the grassland.
The minimum and maximum leaf area index (LAI) were estimated through undercanopy transmittance measurements using the SunScan Type SS1 from Delta-T device, in January and August 2005 under beech trees. In addition, hourly sap flow of six beech trees and five pines was measured during the 2013 growing season with thermal dissipation probes. These were weighted according to species cover and upscaled to an estimate of plot transpiration after sampling trees to assess the sapwood area in the study plot. In addition to the in situ measurements, we used incoming short-wave solar radiation and down-welling long-wave radiation (required as model inputs) from the online ERA-Interim climate reanalysis source (Dee et al., 2011).  FIGURE 3 Summary of the data sets available at the two plots: time series periods and time step resolution. Climate inputs comprise daily time series of incoming short-wave solar radiation and down-welling long-wave radiation; minimum, average and maximum temperature, precipitation, air relative humidity, wind speed The chemical composition of rainfall, forest throughfall (15 rainfall collectors arranged in a cruciform pattern in the forest plot) and soil waters (using suction lysimeters at depths of 30, 50, 70, and 250 cm at the grassland and 30, 50, 120, and 350 cm at the forest) has also been monitored by sampling at approximately biweekly intervals. As an independent check on how well EcH 2 O-iso captures interactions between water storage, flux dynamics, and associated mixing relationships, we used the chloride data collected from precipitation, throughfall and soil water as an assumed conservative tracer in the model for the forest site (cf. Peters & Ratcliffe, 1998). This was not possible at the grassland site as concentrations were too low and uncertainties too high in the absence of monitoring of throughfall and the effects of dry and occult deposition on the grass sward.  (Kuppel et al., 2018b). In this study, we further adapt the model formulation to simulate other passive tracers, such as chloride, which are not subject to evaporative fractionation but evapoconcentration. This flux tracking assumes complete mixing in each storage compartment such that they can be defined by a single average concentration and age, with no preferential age or concentration selection by the water fluxes. Under these assumptions, the concentration and age of outgoing fluxes at each time step correspond to those of the storage compartment at that time.
We used EcH 2 O-iso to model the interlinkages between energy balance, water cycling, and biomass production and quantify the effects on water partitioning and water ages at the Stechlin site. Past studies have successfully applied the model in different environments at the watershed scale (Kuppel et al., 2018a;Lozano-Parra et al., 2014). This study provides an application of the model at the grid-cell scale to resolve and track the age of "green" and "blue" water fluxes.
For the forest plot, measurements of tree biomass production also enabled us to explicitly include in the model calibration process metrics of daily, seasonal, and/or long-term plant physiological dynamics (stem and leaf growth, transpiration, canopy cover, etc.), along with more commonly used observations pertinent to the energy (e.g., soil temperatures) and water balance (e.g., soil moisture) components.
Additionally, we used the chloride time series in precipitation, throughfall and soil water as independent (i.e., not used for calibration) tracers of water fluxes for model verification.
At each time step (daily in this application), the model requires meteorological information of incoming short-wave radiation (R SW ), down-welling long-wave radiation (R LW ), air temperature (maximum T a,max , minimum T a,min , and average T a,mean ), wind speed ( is simulated using the Green and Ampt equation (Mays, 2010). Evaporation is limited to this upper layer. Drainage from the upper soil layers to deeper layers occurs when field capacity is exceeded, the rate of percolation increasing linearly with the water content of the source layer. Leakance through the bottom boundary of the soil follows the same approach, but the rate is further controlled by a leakance parameter to represent the condition range between free drainage and no drainage (i.e., water tight bedrock).
The distribution of soil water losses from transpiration uptake is determined from the fraction of roots present in each soil layer. Simulation of carbon uptake and plant growth is adapted from the 3-PG (Landsberg & Waring, 1997) and TREEDYN3 models (Bossel, 1996;Peng, Liu, Dang, Apps, & Jiang, 2002). Gross and net primary production (GPP and NPP) are calculated using a multiplicative function of the photosynthetically active radiation and the amount of transpired water ( Figure 4c).  (Kuppel et al., 2018a). At the data-rich Stechlin site, we adopted this approach to assess the model skill at reproducing the dynamics of different ecohydrological data sets using long-term, seasonal, and daily periods. We used a range of metrics to determine which parameter sets are "acceptable" or "behavioural" (Beven, 2006;Beven & Binley, 1992). The criteria of the calibration are summarized in Table 2, and the time periods are used for each data set in Figure 3. We used the multicriteria calibration to tune the model at daily time steps using dynamics of soil water content (SWC) at several soil depths and soil surface temperature for both plots. We chose the Kling-Gupta efficiency (KGE) statistic (Gupta et al., 2009) as our metric of "goodness of fit" for time series simulations. This equally considers bias (through assessment of the mean), correlation (assessment of the timing), and variability (assessment of the range of variation). Additionally, measurements of LAI, stem growth, and transpiration rates were used for calibrating the forest plot (Table 2).
We also used various known quantitative thresholds as "observation-driven and expert-knowledge-based constraints" (Kelleher, McGlynn, & Wagener, 2017; Table 2). For example, previous water balance studies of Lake Stechlin (e.g., Pöschke et al., 2018) have For the final multicriteria calibration, we used a Monte Carlo analysis with 30,000 runs. The quantitative assessment using thresholds was used to reject implausible simulations, whereas the dynamic assessment was used to rank the retained parameter sets according to their likelihood. A global score for each simulation (GS i ) was calculated as follows: where LS i v is the variable-specific assessment or "local" score of simulation i (KGE or RMSE; see Table 2) and σ (LS v ) the standard deviation of LS v over the 30,000 runs. The latter standardization aimed to derive a simple weight balancing of the variable-specific assessments when calculating the GS. The GS was finally used to select the 15 best simulations out of the likelihood simulations.

| Parameterization and sensitivity
As a preliminary step to model calibration, an analysis of parameter sensitivity was conducted to reduce the number of free parameters: any parameter, showing neither first nor second order sensitivity related to one of the LS, was removed from calibration parameter set and fixed to a value according to locally measured properties or to literature values.  An exception is the SWC of the third layer of the grassland plot where measured variability is low (see below and Figure 2).
When the model is calibrated against all data sets simultaneously (global calibration), there is an expected overall decrease in the model performance compared with the performance achieved by individual outputs calibrated specifically using time series directly related to such output (e.g., when modelled soil moisture is calibrated using soil moisture observations; Figure 5). The decreased performance was generally greater in the forest site, though the performance depreciation was small or nonsignificant (at 95% confidence) for SWC1, SWC2, SWC3, temperature differences, and transpiration rate (in general ΔRMSE < 11%; ΔMAE < 8%; ΔPearson < 0.15). LAI and biomass production (ΔMAE = 20% and 17%, respectively) clearly had a higher deg-

| Simulation of energy, hydrological, and biomass components
The short term and seasonal dynamics of soil surface temperatures were generally represented with good accuracy for the forest and grassland plots, though there was a slight overestimation at both plots during the early growing season, and an underestimation at the forest plot in late summer (Figure 6a  which is slightly high for temperate grasslands and may overestimate interception (Munier et al., 2018). In the forest plot, the high LAI increase during the wet summer of 2007, and the unrealistic limited decline during the winter seasons was also notable, which seems to reflect excessive allocation to leaves when moisture is not limiting.
This is not consistent with the direct measurement of leaf and needle fall (Table 3) nor with the observed seasonal LAI dynamic (Figure 8), which are unrelated to the variable climatic conditions. This is likely explained by the high sensitivity of biomass allocation in the model to many parameters and may account for why there is a trade-off in the global calibration between biomass production and LAI dynamics.
The average modelled above-ground forest biomass production underestimates measured values by~10-30% (Table 3). However, results are broadly consistent between years, and the simulation confidence interval brackets the observation values ( Figure S1). Notice that leaf allocation is very consistent between years; however, observation of fruit production shows high variability, suggesting that the stem and fruit pool might be a buffer in the model or capture phenological interannual dynamics unrelated to climate (Lebourgeois et al., 2018), and these detailed physiological aspects are not captured by the model.

| Water balance for forest and grass plots
Averaged over 11 years, the most marked modelled water balance differences between the two sites were clearest in the higher interception losses from the forest site (though the high LAI and often low intensity rainfall still results in substantial grassland interception; Table 4). Transpiration and evaporation losses were also higher, though the differences with the grassland were smaller. Consequently, average groundwater recharge under the forest was roughly a third of that under grassland (73 mm compared with 225 mm), mostly due to the trade-off with interception and higher transpiration. Figure 10 shows     the winter months, but unlike in the forest plot, it can also be high in the summer.
Although the model seems to be successfully partitioning overall "green" and "blue" water fluxes, there is significant uncertainty regarding the allocation of soil evaporation and transpiration in the forest. As a result, forest soil evaporation seems too high, especially in the early spring. Evaporation is sensitive to the trajectory of the modelled LAI, as the slow increase in early growing season allows energy to reach the forest floor and drive evaporative cooling, which is further exacerbated by the cooling effect of the soil temperature as a calibration constraint.
In the grassland plot, groundwater recharge exhibits the greatest interannual variability (standard deviation of +105 mm), with low variation in modelled interception, transpiration, and soil evaporation rates (Table 4). Although groundwater recharge also has the greatest interannual variability in the forest plot (standard deviation +67 mm), variability in transpiration (+55 mm) is similar and high, depending upon the balance between soil moisture availability and atmospheric demand.  Figure 11). High winter chloride inputs propagate rapidly through the soil profile advecting with the wetting fronts at 30 and 50 cm ( Figure S4a). In spring and summer, concentrations decline in precipitation and also in the soils, despite evapoconcentration. At depth (120 cm), there was a usually slight lag of a few weeks in the winter peak of Cl and more attenuation of the decrease in concentrations ( Figure S4a). However, in general, average concentrations increased with depth as the effects of evaporation and transpiration were apparent. This was difference was particularly marked in summer ( Figure S4b).

| Estimation of ages of fluxes
The dynamics of water flux ages were broadly similar for both plots ( Figure 12), though lower water fluxes beneath the tree canopy resulted in greater ageing of water at depth in the forest (Table 5).
The age of evaporated water varied between 1 and 90 days with a mean of around 45 days for both sites. Transpired water was oldest in the forest site, with a mean of 176 days, compared with 87 days in the grassland, reflecting the age of deeper water sources tapped by the influence of greater rooting depths in the distribution calibrated for the forest plot. Generally, ages in the different pools were moreor-less strongly influenced by rainfall distribution at both plots. For the top layer, there was a direct flushing effect on the modelled ages, which declined during and after major rainfall episodes. In the second and third layer, there was significant decrease in water age, mostly when storm events storms were particularly large (e.g., in the summers  . Tracer-aided ecohydrological models such as EcH 2 O-iso have the potential to enhance this capability by providing a means of conceptualizing the mixing that occurs in water storage -flux interactions, the effect on water ages, and to help constrain the sources of evaporation and plant water use (Kuppel et al., 2018a). By integrating energy exchanges, water fluxes and biomass dynamics, such tracer-aided models also have the potential to provide quantitative insights in to how land cover regulates the interlinkages between water storage-flux-age at different spatial and temporal scales.
Ecohydrologic models tend to be highly parameterized, and the opportunity for insight only exists if a limited number of feasible and consistent model configurations can be identified. Multicriteria calibration and verification of such models can increase the confidence that the dominant ecohydrological processes are being appropriately represented in different landscape compartments and accurately quantified (Kelleher et al., 2017;Kuppel et al., 2018a). Model failure to adequately represent observed processes also provides an opportunity to learn and improve conceptualization Dunn et al., 2008). The application of EcH 2 O-iso to the monitoring site at Lake Stechlin, followed on from a successful catchment-scale application of the model in a wet, boreal catchment in Scottish Highlands (Kuppel et al., 2018a), and the present study provided an opportunity to test the model in a comparative forest/grassland plot-scale study in a more water-limited site and, more importantly, use direct

| Water balance implications
The model simulated a balance of "green" and "blue" water fluxes at the forest site and groundwater recharge rates that were consistent with other investigations at Stechlin (e.g., Pöschke et al., 2018).
Although soil moisture had a general seasonality related to winter wetting and summer drying, the permeable nature of the soils along with the occurrence of wet summers with occasionally heavy rainfall, means that recharge can occur all-year round, particularly in the grassland site, and this was well represented by the model. age is also reflected in the simulations of the chloride concentration.
The performance of the chloride concentration simulations is best in in the upper soil layer, indicating that evapotranspiration fluxes are accurately quantified and that full mixing of the incoming tracer signal is reasonably approximated. However, in L2, the uncertainty over the volume and depth of transpiration explains why the model captures the general seasonality of the chloride signal reasonably well but does not reproduce the concentrations accurately. However, it should be noted that soil water was sampled by lysimeters under low tension and may not reflect the evaporative signal in finer soil pore waters (Geris, Tetzlaff, McDonnell, & Soulsby, 2017). In contrast, the conceptualization of mixing in EcH 2 O-iso tracks the age and tracer concentration of bulk soil water, which includes the water held under higher soil tensions. The effect of this difference may be smaller in the sandy soils of the study site, but even when dry, the residual volumetric SWC in these soils is still 5% and can contribute to the difference between the concentrations in the pool of soil water that is sampled and the concentrations in the bulk soil water represented by the model.
Although the celerity of wetting front moving through the soils was evident in the observations and the simulations, the mixing appears to attenuates this modelled signal too rapidly in the deeper soil layers (L2 and L3). A better differentiation of faster and slower flowing water in the model and partial mixing between them would likely improve the chloride simulations (e.g., Sprenger, Tetzlaff, Buttle, Carey, et al., 2018).

| Implication for using tracer-aided ecohydrological models in land use change studies
Working towards enhancing ecohydrological models through the use of tracers to improve the representation of subsurface mixing and collecting informative data of different types to improve model calibration and parameterization are important goals to refine our models and assess the hydrological implications of land use change under projected climate change scenarios. However, advancing these agenda is predicated on the idea that we know what aspects of the model need to be improved and what data collection efforts should be prioritized. Models can inform these aspects (Peters, Freer, & Beven, 2003).
For instance, the excessive damping of the chloride signal indicates that our model needs to incorporate partial or differentiated mixing of faster and slower zones of water movement in the soil (Sprenger, Tetzlaff, Buttle, Carey, et al., 2018). This also provides a basis for better apportioning of the pools of soil water that contribute to evaporation and transpiration and permits a more meaningful comparison with the water compositions sampled by lysimeters.
In addition to better tracer data, improved ecological data sets that represent seasonal to longer term processes are critical to improve the calibration of model components that simulate changes in plant biomass with critical hydrological feedbacks at interannual to decadal timescales. Acquisition of biomass data during measurement campaigns has only recently become common in the hydrologic community. Measurements of the forest LAI over the growing season are probably one of the most important measurements and are relatively common, but additional measurements on root distribution, stem growth, or total biomass are also critical and less common.
As in many modelling studies, this investigation leverages data sets collected for other purposes. Often these data sets provide key longterm historical information; however, they are often not optimal to inform models because the campaigns were not been specifically designed for this purpose. It is clear that as the fields of ecohydrological and tracer-aided modelling mature, careful planning of data acquisition to best enhance model development and testing is necessary. In the future, such coevolution may facilitate more rapid advances in our understanding and ability to accurately predict the hydrological impacts of land use change. This is a priority for work planned in the drought-sensitive study area of Brandenburg, where the effects of climate change, and predicted warmer, drier summers, may result in scarcity of water resources in future (Lischeid & Nathkin, 2011). More accurate ecohydrological modelling is needed to inform decision making on how different land use scenarios, in terms of the balance of forest and nonforest cover, will involve trade-offs in terms of water availability and other ecosystem services.

| SUMMARY AND CONCLUSION
We applied a process-based ecohydrological model (EcH 2 O-iso) to compare the effects of land cover on water partitioning and associated flux ages under temperate beech forest and grassland on podzolic, sandy soils at Lake Stechlin in Germany. Multicriteria calibration, based on measures of the energy balance, hydrological function, and biomass accumulation, resulted in generally good simulations of surface energy exchange, SWC, transpiration, and biomass production.
The model results showed that the forest used more water than the grassland from the 620 mm of average annual precipitation. On average, "green" water fluxes from interception, transpiration, and evaporation were 88% of precipitation inputs under beech forest, compared with 63% under grassland. As a result, groundwater recharge was greatly enhanced under grassland at 37% of precipitation compared with 12% for forest. The model also tracked the ages of water in the different fluxes. In shallow soil horizons, the average ages of soil water fluxes and evaporation were similar in both plots (~1.5 months), though transpiration and groundwater recharge were older under forest (~6 months compared with 3 months for transpiration and~12 months compared with~10 months for groundwater).
Flux tracking with Cl tracers provided independent support for the modelling results, though it also highlighted effects of uncertainties in the model. To realize the potential for tracer-aided ecohydrological models in land use change studies, further improvements in the conceptualization of soil water mixing and carefully planned data acquisition on biomass dynamics seems the highest priorities for more reliable predictions.

ACKNOWLEDGMENTS
The authors would like to acknowledge research funding from the European Research Council (project GA 335910 VeWa). M. Maneta acknowledges support from the U.S. National Science Foundation (project GSS 1461576). C. S. is grateful to the Leibniz IGB Berlin for a Senior Research Fellowship. We also thank Umweltbundesamt (UBA) for providing the climate data.

DATA AVAILABILITY STATEMENT
The data that support the findings of this study are available from the corresponding author upon reasonable request.