Meteorological data including air temperature (∘C), relative humidity (%), rainfall or snowmelt amount (mm day-1) (Sprenger et al., 2018b), and potential evapotranspiration (mm day-1) estimated using the Penman–Monteith equation (Allen et al., 1998) were available on a daily basis at each catchment.

Soil hydraulic characteristics, as shown in the Supplement (Fig. S2), were derived for Bruntland Burn and Dorset from the pedotransfer functions provided by Schaap et al. (2001) using site-specific soil textural and bulk density information (Sprenger et al., 2018b). For Krycklan, the hydraulic parameters were estimated based on laboratory measurements (Nyberg et al., 2001).

The simulations of travel times and water ages are based on tracking, in a 1-D soil hydraulic model, the infiltrated water (rainfall and snowmelt) with a virtual tracer in soil storage and fluxes leaving the soil. We applied the SWIS model as described in detail by Sprenger et al. (2018b). The SWIS model solves the Richards equation for water flow and simulates tracer transport with the advection–dispersion equation. The SWIS model can partition the subsurface into two flow domains (Fig. S3): a fast flow domain representing the soil pores that hold the water at pressure heads < 600 hPa and a slow flow domain covering the pores with a water retention > 600 hPa. A threshold of 600 hPa was chosen to divide the two pore domains, as this is approximately the pressure head applied by suction lysimeters to extract water. This definition allowed us to use stable isotope data (2H and 18O) of the mobile flow domain (sampled with suction lysimeters) and the bulk soil water (slow plus fast flow domain) sampled with the direct equilibration method (Wassenaar et al., 2008) for benchmarking the model performance at the individual study sites as presented by Sprenger et al. (2018b). Ingraham and Criss (1993) found that two water pools approach as a function of water volumes, surface area, and saturated vapour pressure (temperature) a weighted average isotopic composition of the two pools. Our previous study showed that a conceptualization of the subsurface with two pore domains that exchange water in accordance with Ingraham and Criss (1993) via the soil gas phase improved the simulation of the soil water stable isotopic composition at 10 and 20 cm depths at the investigated sites compared to an assumption of uniform flow. Therefore, we apply the same model set-up of SWIS as presented in detail by Sprenger et al. (2018b) with the parameters given in Table 1. In accordance with Vanderborght and Vereecken (2007), we set the dispersivity parameter to 10 cm at all sites. The soil physical parameters were the same for the two pore domains and the exchange was solely conceptualized as vapour exchange rather than via hydraulic dispersion. The implemented tracer exchange between the slow and fast flow domains results in a slow approach of the virtual tracer concentrations in the two pore domains. Thus, the exchange leads towards a homogenization of water ages between the two flow domains. Consistent with soil physics principles, the slow flow domain is filled first and remains saturated until the fast flow domain is emptied (Hutson and Wagenet, 1995). Water flow and tracer transport occur in both domains and recharge is generated accordingly. However, only the combined recharge flux rate (R) and weighted average tracer concentrations from both domains are provided. The model domain covered the soil profile down to 50 cm depth in 5 cm intervals. Root water uptake was limited according to rooting depth observations to the upper 15 cm at the heather site in Bruntland Burn and to the entire 50 cm soil profile at the forested sites. Soil evaporation (E) was limited to the upper 10 cm based on experiments by Or et al. (2013). ET was partitioned into potential E and potential transpiration (T) according to the canopy coverage (Table 1) according to Ritchie (1972). Since sap flow was measured at the forest site in Bruntland Burn (Wang et al., 2017a) and E estimates based on the maximum entropy theory were available for the heather site in Bruntland Burn (Wang et al., 2017b), we used this information to adjust the partitioning of ET at these sites. E and T both decreased linearly with depth and occurred from both the fast and slow flow domains (T limited to the permanent wilting point assumed to be at 15 000 hPa, Fig. S2). Contrary to the application of the SWIS model for stable isotope modelling, E did not alter the virtual tracer concentration (similar to T), but reduced the soil moisture at the depths of E losses and root water uptake, respectively. Precipitation was divided into interception and throughfall according to the canopy coverage (Table 1), and when the interception capacity (Table 1) was reached, the surplus infiltrated into the soil. Soil frost does usually not occur at Bruntland Burn and is rare at the Dorset site due to the insulating effect of the snow cover. At Krycklan, soil frost was shown to not induce surface runoff, but soils at the forested site remained permeable (Stähli et al., 2001; Laudon et al., 2007).

We defined the start of travel times and zero water age of waters as the day of infiltration into the soil profile. To derive the travel times and water ages, we ran the SWIS model for each day of rainfall or snowmelt from 06/2011 for Bruntland Burn and Dorset and from 01/2010 for Krycklan to 09/2016 and tracked the fate of a virtual tracer in soil storage (fast flow domain, slow flow domain, and total storage) and water fluxes (E, T, and R) as suggested by Sprenger et al. (2016). The number of days with rainfall or snowmelt of all days of simulation were 1381/1943 for the Bruntland Burn sites, 684/1984 for Dorset, and 801/2465 for the Krycklan site. The model was run at daily resolution and the applied parameters are listed in Table 1.

Consistent with the definitions by Benettin et al. (2015), we consider two different metrics as conceptualized in Fig. 1. The first was the median travel time (MdTT) as a forward approach that estimates how long it takes the infiltrated water to leave the soil as E, T, or R flux (also called life expectancy in Benettin et al., 2015). The second was the median water age (A) as a backward approach estimating the age of water in the output fluxes and the soil storage since it infiltrated into the soil (also called residence time in Benettin et al., 2015).

To derive the MdTT, we extracted the breakthrough curves of the normalized mass fluxes Oj(t) in the output fluxes (E, T, and R) generated from each virtual tracer mass introduced during individual infiltration events (Ij) on day j (Fig. 1, left). Normalized mass fluxes Oj(t) (T-1) resulted from the tracer concentration (introduced with Ij) in the flux Cj(t) (M⋅ L-3) times flux rate Q(t) (L3⋅T-1) divided by the introduced tracer mass Ij (M): Ojt=Cjt⋅QtIjt=0. We then computed the median of the individual breakthrough curves as half of the maximum cumulative Oj(t), which then described the time it took until 50 % of the infiltrated water ended up in the considered output flux from the soil (Sprenger et al., 2016). This leads to a time series showing the median travel times (MdTT in days) required to leave the system via either evaporation (MdTTE), transpiration (MdTTT), or recharge (MdTTR). Since MdTT would be underestimated if not all of the virtual tracer had left entirely the soil storage, we limited the MdTT analysis to the period from 2012 to 2015. We decided to present median values, rather than mean travel times, as the latter can be biased due to uncertainties in the long tails of the transit time distributions (Seeger and Weiler, 2014). We further used the individual breakthrough curves in the E, T, and R fluxes to derive master travel time distributions (MTTD) as introduced by Heidbüchel et al. (2012). In line with Heidbüchel et al. (2012), we superimposed all individual breakthrough curves, weighted them by the event size, and normalized them by the total introduced virtual tracer mass. Such a weighted average travel time distribution was derived for evaporation (MTTDE), transpiration (MTTDT), and recharge (MTTDR) fluxes at each study site. The time after which 50 % of the average tracer mass has left via the considered flux was defined as the median of the MTTD.

We calculated water ages A(ti) by first multiplying Oj(t) (T-1) by the precipitation amount Pj(t= 0) (L3) that introduced the virtual tracer Ij(t= 0) and divided it by the flux volume Q(ti) on the day that we estimated the water ages for, to get the relative share of each precipitation event introduced on day tj in the considered fluxes Vj(t) (T-1). Vj(t)=Ojt⋅PtjQ(ti) Multiplication of Vj by the days since tj provides the relative volume of the water of age ti-tj for each considered day (Fig. 1, right). The 50th percentile of the cumulative sum of Vj then defined the median water age. To prevent bias due to water of unknown age in the soil storage (i.e. initial water in the soil at start of simulation), we limited the water age analysis to the period from 2013 to 2016. Here, we report the median water ages in the fast flow domain (ASf), the slow flow domain (ASs), and total soil storage (ASt), and of the evaporation (AE), transpiration (AT), and recharge fluxes (AR).

Distributions of the time-variant median travel times and median water ages in fluxes and storages were derived using cosine kernel density estimations (Venables and Ripley, 2011). Differences in MdTT and water ages between the four sites were analysed using the non-parametric Kruskal–Wallis test with a post hoc Dunn test (significance level of 0.05), since the data were not normally distributed according to the Shapiro–Wilk test. Since running the model for the considered 5 years took between 1 and 3 h and our analysis required running the model between 684 and 1381 times (for each day of precipitation), we were not able to do a formal uncertainty analysis due to the long computation times (up to > 100 days for one set of MdTT and water ages per site). We therefore limit the presented water age analysis to one realization using a parameter set that was previously shown to reflect the water flow and transport dynamics well using stable isotope data (Sprenger et al., 2018b).

As the age of the total soil water as well that of the fast and slow flow domains (ASt, ASf, and ASs, respectively) generally decreased with increasing storage volume, the antecedent hydro-meteorological conditions controlled the soil water age dynamics. For periods with high ET fluxes that led to low storage volumes, water ages in the soil pores generally increased over time (Bruntland Burn sites in Fig. 6), because the youngest water left the soil column preferentially via ET. Thus, E rates and vegetation uptake directly impacted water age dynamics in the critical zone. Additionally, snowmelt led to a sharp decrease in soil water ages after a continuous aging of the water that resided in soil over the snow accumulation period (Fig. 5). Both the ET-driven and snowmelt-driven cases result in an inverse storage effect, where water in the soil became younger for higher soil water volumes. In addition to the general positive relationship between wetness and soil hydraulic conductivity (van Genuchten, 1980), the conceptualization with two pore domains in the SWIS model allowed young water in the fast flow domain to bypass older water stored in the slow flow domain. Since the smaller pores of the slow flow domain will be filled first or stay filled while the larger pores of the fast flow domain are not empty, the bypass will be enhanced during periods of high wetness. As a result, young median water ages prevailed across the entire 50 cm soil profile during periods of high storage (Fig. 7). Note that this conceptualization would not hold when soil dryness induces preferential flow due to water repellency (hydrophobicity) (Ritsema et al., 1993; Weiler and Naef, 2003).

According to our simulations, water ages are not simply controlled by the hydraulic conductivity of the soil, but the storage dynamics in the slow and fast domain also impacted the water age dynamics. Water flow was much slower when the fast flow domain emptied. Consequently, while the hydraulic conductivities at the Bruntland Burn and Dorset sites were similar (Table 1), the water ages at the two forested sites, where the fast flow domain dried out during summer (Fig. S4), were greater than at the heather site at Bruntland Burn, where water prevailed in the fast flow domain throughout the year. The impact of increased water mobility (i.e. flushing) on soil water ages during greater soil wetness is supported by stable isotope data (2H and 18O). For the Bruntland Burn sites, Sprenger et al. (2017) showed that the isotopic variability in bulk soil water was greatest after intense infiltration events, revealing that event water mixed effectively with pre-event water in the upper 20 cm. Also, isotope data from mobile soil water (Peralta-Tapia et al., 2015) and bulk soil water (Sprenger et al., 2018a) at the Krycklan site had a strong relationship with the isotopic compositions of previously infiltrated water, which shows that a high proportion of the soil (pre-event) water is replaced by or mixes with recent event water. Such isotope studies provide a snapshot view of water transport in the field, but modelling approaches benchmarked against or calibrated on such observations, like our study, allow insights into short-term dynamics.

ASs (and the storage volume in the slow flow domain) generally does not change as rapidly as ASf, which is influenced by highly variable storage volumes in the fast flow domain. However, since the ratio between water stored in the fast and slow flow domain changes as a function of soil wetness (Sprenger et al., 2018b), the impact of the two domains on total soil water age ASt also varies over time. During summer, when storage in the fast flow domain decreased or was even fully depleted (at forested Bruntland Burn and Dorset sites), ASt approached or equalled ASs. Due to exchange between fast and slow flow domain, ASf approached ASs just before the fast flow domain dried out (see forested Bruntland Burn and Dorset in Fig. 5c). Our age analyses therefore support the hypothesis by Sprenger et al. (2017) that old water in smaller soil pores can lead to a “memory effect” in the bulk soil water isotope compositions. Such a “memory effect” was further shown to lead to a lagged response of the soil water stable isotope compositions to hydro-meteorological forcing at five long-term experimental catchments in northern environments (Sprenger et al., 2018a). Further, observed differences in the isotopic compositions of mobile and bulk soil water in the field were often related to the potential age differences of waters sampled at different mobilities (Landon et al., 1999; Brooks et al., 2010; Geris et al., 2015; Sprenger et al., 2015a; Oerter and Bowen, 2017). Our results and recent simulations by Smith et al. (2018) support such interpretations, as the water in the slow flow domain was generally older than the water in the fast flow domain (ASs > ASf). However, since the differences between ASs and ASf were variable in time and were often maximized in early spring, such anomalies in water ages are likely to be reflected in the isotopic compositions of the water, with the older water in small pores being less depleted in heavy isotopes (originating partly from autumn precipitation) than the young water in larger soil pores draining recently infiltrated isotopically depleted snowmelt or winter precipitation. Such isotopic differences resulting from different water ages affect our interpretation of soil water stable isotopes sampled either with suction lysimeter (mobile water in the fast flow domain) or cryogenic vacuum extraction (bulk soil water in fast and slow flow domain). For example, Brooks et al. (2010) reported different isotopic compositions for mobile and bulk soil water samples, which led to the formulation of the two water world hypothesis (TWW) (McDonnell, 2014). In a TWW scenario, tightly bound soil water is not displaced via translatory flow, does not mix with or displace mobile water, and does not enter the stream. However, experimental work recently showed that there is interaction between mobile and less mobile soil waters (Vargas et al., 2017), as conceptualized in the applied SWIS model. Our simulations further question if water in the slow flow domain – as defined in SWIS – will not eventually recharge the groundwater and streams. The virtually introduced tracer eventually disappears from the soil water storage of the slow flow domain, due to loss of the tracer to the atmosphere (ET flux), interaction with the fast flow domain and recharge within the slow flow domain. Nevertheless, while definition of the slow flow domain using a higher threshold pressure head (e.g. field capacity as suggested by Brantley et al. (2017) rather than the currently assumed 600 hPa) would result in its water becoming more tightly bound, interaction with more mobile waters would likely still persist (Vargas et al., 2017).

Since water loss from soil storage as E flux was limited to the upper 10 cm in our simulations, travel times and water ages are directly related to the water age dynamics in the top soil. In contrast, the rooting zone covered the entire soil profile for the forested sites and down to 15 cm soil depth for the heather site, which affected the resulting travel times and water ages of T accordingly.

Investigation of water ages in the ET flux is relatively new (Botter et al., 2011) and age dynamics have usually been assessed for the bulk ET flux (Harman, 2015; Soulsby et al., 2015; van der Velde et al., 2015; van Huijgevoort et al., 2016; Soulsby et al., 2016). While it was shown that tracer-aided modelling using stable isotopes of water benefits from partitioning ET into a fractionating E flux and a non-fractionating T flux (Knighton et al., 2017), separate water age analyses for E and T have been considered only recently (Smith et al., 2018). However, our analyses showed that the two different fluxes can have markedly different travel time dynamics (Fig. 2), average travel time distributions (Fig. 4), and water age dynamics (Fig. 8). Thus, our process understanding of how vegetation affects water ages in hydrological systems would particularly benefit from further assessments of the differences between E and T water ages. Such investigations are of special interest in light of ongoing research regarding the consequences of a potential TWW hypothesis on water age estimations based on tracer-aided modelling (Hrachowitz et al., 2016). In particular, our simulations underline that neither is ET flux withdrawal well mixed in its age composition nor is the pool of plant water uptake well mixed, which is increasingly acknowledged in water age studies (Harman, 2015; Smith et al., 2018).

Similar to findings by Smith et al. (2018) for the heather site in Bruntland Burn, our estimates for AE at that site were highest during periods of limited infiltration (e.g. 10-year return period drought in summer 2013 at Bruntland Burn in Fig. 8a). However, E water ages reported by Smith et al. (2018) were higher than our estimates, which is probably due to their conceptualization of the subsurface into one domain with and one without downward flux, which resulted in generally higher water ages in the shallow soils compared to our estimates. Water age estimates by Queloz et al. (2015) for the ET flux from a lysimeter were less variable, but within about 10 to 20 days of the magnitude of our AE and AT estimates. The forward travel time distributions for water leaving the soil via ET presented by Queloz et al. (2015) also showed shapes similar to our reported MTTDE and MTTDT with peaks in the first few days and tails of the distribution that can reach up to 200 days (Fig. 4). We attribute the long tails of the MTTDE and MTTDT to both the ET flux from the slow flow domain and root water uptake from deeper soil layers.

With regard to AT, our soil physical model showed a similar inverse storage effect as the approach using storage selection functions (Smith et al., 2018): water taken up by plants was generally younger during higher soil storage. While Smith et al. (2018) had a dynamic root water uptake depth, T loss in the SWIS model decreases linearly with depth as long as the pressure head does not reach the permanent wilting point, which is usually not reached at the investigated sites (Sprenger et al., 2018b). Thus, it is likely that the differences between the dynamic root water uptake depths in the storage selection functions and the defined uptake profile in SWIS will be more pronounced when vegetation responds to intense drought by shifting the root water uptake to deeper soil layers (Volkmann et al., 2016).

A relationship between MdTTT and T dynamics, with the onset and cessation of T at the beginning and the end of the growing season, has been shown previously (Sprenger et al., 2016); nevertheless, our experimental set-up with two different vegetation types (differing in T rates, rooting depth, canopy cover, and interception storage) on similar soil types under the same climatic forcing in the Bruntland Burn reveals the impact of rooting depth on the travel time dynamics. Median and maximum MdTTT were shorter for the heather site than for the forested site in Bruntland Burn and the MTTDT had substantially different shapes at both sites with a lower median for the heather T travel times compared to the forest (Table 1).

Water age and travel time dynamics of the recharge flux are the result of the interplay between the aforementioned linkages between soil water storage age and ET age dynamics. Since the estimated MdTTR are a function of the subsequent recharge flux intensities (Fig. 3), the probability of an introduced water parcel leaving the soil profile via recharge is higher during high flows. Such a relationship between forward travel times and the recharge flux dynamics was also found in modelling studies on a controlled lysimeter (Queloz et al., 2015) and 35 field sites in Luxembourg (Sprenger et al., 2016). While catchment-scale travel time studies based on conceptual lumped models also showed that the subsequent precipitation patterns affect the travel time dynamics of runoff (Heidbüchel et al., 2013; Hrachowitz et al., 2013; Harman and Kim, 2014; Peters et al., 2014; Peralta-Tapia et al., 2016), our application of a 1-D soil physical model provided insights into the processes in the upper critical zone leading to such behaviour at the plot scale. The simulations with SWIS highlight the effect of ET fluxes on recharge travel times, as both storage and recharge are influenced by ET rates. Consequently, one can see the exceptionally high MdTTR for the few infiltration events during a 10-year return period dry episode in summer 2013 for the heather site at Bruntland Burn. Further, the seasonal decrease in MdTTR due the preferential recharge during the dormant season (Bruntland Burn) or snowmelt (Dorset and Krycklan) emphasizes the impact of ET on vadose zone travel times. Such an influence of vegetation on travel times as suggested from the plot-scale simulations is commonly not seen for the catchment runoff as the stream integrates water moving via different pathways and thus obscures any ET signal (Tetzlaff et al., 2014; Kirchner, 2016).

The conceptualization of fast and slow flow domains resulted in MTTDR that indicated a maximum probability of infiltrating water recharging from the soil within 3 to 10 days after infiltration, although the tails of the MTTDR revealed that replacement of all water (turnover time) can take up to 1000 days (Fig. 4). Thus, our modelling approach of a two-pore domain enabled the representation of the short-term responses and the long-term memory of the recharge composition in a soil column. As a result, water ages (ASt) did not always increase with depth, but instead became almost uniform throughout the soil column during intense infiltration periods. Occasionally ASt was smaller at the bottom of the profile relative to just below the rooting zone, mainly due to root water uptake dynamics at the heather site in Bruntland Burn (Fig. 7). While AR was generally higher than ASt, consistent with Queloz et al. (2015), our soil physical modelling approach revealed how the water ages develop with depth and lead to the resulting AR dynamics.

The pronounced longer water ages of the slow compared to the fast flow domain are of great relevance for the interpretation of studies on travel times in vadose zone water fluxes. These investigations are often based on models calibrated with isotope data from samples taken with zero-tension lysimeters (e.g. Asano et al., 2002), wick samplers (e.g. Timbe et al., 2014), suction lysimeters (e.g. Muñoz-Villers and McDonnell, 2012; Tetzlaff et al., 2014; Hu et al., 2015), or from the outflow of lysimeters (e.g. Stumpp et al., 2009, 2012). Such methods limit isotope sampling to the most mobile water in the soil (Sprenger et al., 2015a), which is represented as the fast flow domain in the current application of the SWIS model. According to our simulations, travel time studies based on the most mobile waters in the soil are likely to underestimate travel times and water ages in the recharge fluxes. Consequently, the turnover time of the soil pores will be underestimated in such studies, which can then lead to the assumption that nutrients or contaminants located in the vadose zone will be flushed out more rapidly than they actually are.

The R water ages at 50 cm depth in this study are obviously younger than catchment-scale runoff water ages (Soulsby et al., 2015; Ala-aho et al., 2017b; Benettin et al., 2017; Kuppel et al., 2018; Piovano et al., 2018). Nevertheless, the dynamics from the plot-scale R water ages are similar to the catchment runoff water ages due to hydro-meteorological controls leading to generally increasing values throughout spring towards summer (decreasing storage) and the lowest water ages during winter (highest storage).

The long tails of the MTTDR found in our results indicate that soil storage can probably add to the commonly observed long tails of catchment-scale travel time distributions (e.g. Godsey et al., 2010; Hrachowitz et al., 2010). Generally, the plot-scale soil hydraulic simulations can help to better understand the processes taking place within the catchment and to constrain or benchmark spatially distributed hydrological models (van Huijgevoort et al., 2016; Ala-aho et al., 2017b; Kuppel et al., 2018). Such catchment models cannot account for the soil physical processes in a similar detail to a 1-D model due to computational limitations. However, our results imply that it might be worth adding a dual-porosity representation, similar to the conceptualization in SWIS, to the recently published EcH2O-iso (Kuppel et al., 2018).

While we cannot provide uncertainty estimates for the presented travel times and water ages due to restrictions imposed by computation time, comparison with soil moisture and stable isotope data at each site (Sprenger et al., 2018b) indicates that the SWIS model captures the water flow and transport processes well. However, model calibration using soil moisture and stable isotope data, as suggested by Sprenger et al. (2015b), would supply the basis of an assessment of how different parameter sets impact the model performance and water age estimates. Such an approach would provide site-specific characterization of the soil physical properties and would likely improve simulations compared to the currently applied pedotransfer functions and measurements on soil cores.

The applied model approach cannot account for preferential flow, but the conceptualization of two pore domains with different water flow and transport dynamics enabled the simulation of bypass flow. This conceptualization was shown to be superior to a conceptualization of a uniform flow (Sprenger et al., 2018b). Additional inclusion of preferential flow in the model domain would come at the cost of model complexity and pose problems of parameter identifiability.

Since the investigated northern environments seldom experience severe drought, plant growth is usually not water limited, as for example was shown by Wang et al. (2017a) for Bruntland Burn. Thus, the assumption of a linear decrease in root water uptake with depth appears to be reasonable for the current study sites. An exponential distribution would not change the water uptake patterns significantly, as the linear assumption already results in 96 % of the water being taken up in the upper 15 cm. However, several isotope studies have shown that the root water uptake profile does not coincide with the root distribution if plants experience water stress (e.g. Kulmatiski and Beard, 2013; Ellsworth and Sternberg, 2015; Volkmann et al., 2016). Hydraulic lift can further increase the complexity of soil–plant interaction, as experimentally observed and implemented in a soil hydraulic model by Meunier et al. (2018). Thus, an improved representation of such dynamics in the water uptake depths would be beneficial for modelling studies in arid environments. For the northern environments considered here, variability in the depths from where the plants take up their water appears to be limited (Smith et al., 2018).

The presented simulations are further limited by the discretization of the soil profile into 5 cm intervals, as this might be too coarse for an adequate representation of the interactions between atmospheric demand, T losses, and mixing down the profile. However, computational limitations did not allow a smaller discretization and our study aimed to test the assumption that soil storage is a well-mixed water source of ET fluxes.

Lastly, while the investigated sites are not located on steep slopes, the 1-D simulation cannot account for lateral flows in the vadose zone that may potentially occur during extreme rain events (Soulsby et al., 2017). We further have to assume that all R flux leaving the soil profile will end up in groundwater, but spatially distributed catchment models (Ala-aho et al., 2017b; Kuppel et al., 2018) might reveal that such water could end up in the ET flux from saturated areas in the valley bottom when the groundwater feeds the riparian zone.