Questions such as “What are the preferential water sources for plant root uptake?”, “To which extent do soil physical properties (e.g. soil texture, percentage pore space, soil organic matter content) influence the isotopic composition of water taken up by plants?” and “What are the implications of different transport times and water storage within the root zone and within plants?” are central to understanding water fluxes in the critical zone, and they have been explored in many stable isotope studies (e.g. White et al., 1985; Dawson and Ehleringer, 1991; Stahl et al., 2013; Bowling et al., 2017; Evaristo and McDonnell, 2017). While isotope measurements have become more accurate over the years and progress has been made in quantifying the proportions of different water sources by using Bayesian mixing models (such as SIAR, MixSIAR; see, for example, Evaristo et al., 2017; Rothfuss and Javaux, 2017), many conceptual and methodological challenges remain.

For example, some studies conducted at the catchment scale have found that water taken up by roots was isotopically different from stream water and groundwater (Brooks et al., 2010; McCutcheon et al., 2017). This suggests either that source-water sampling was incomplete, that fractionation processes modified the isotope composition of the water taken up by plants before or during uptake itself, or that other methodological issues may limit the utility of stable isotopes in tracing ecosystem water fluxes. Vital here is the issue of how to appropriately sample and extract water from soil and plant tissues. Several studies have shown that water from the same soil or xylem sample can have different isotopic compositions when extracted with different techniques. For example, soil water extracted with tension lysimeters may be isotopically different from that extracted with cryogenic distillation (Landon et al., 1999; Koeniger et al., 2011; Orlowski et al., 2016b; Gaj et al., 2017b; Thoma et al., 2018) and plant water extracted cryogenically may differ from water directly sampled from the xylem (e.g. Volkmann et al., 2016b; Zhao et al., 2016). Isotope ratios may also differ with different extraction times (West et al., 2006). It has been suggested that these differing signatures may represent different fractions of the total soil- or plant-water reservoir. What is lacking, but urgently needed (Berry et al., 2017), is to (i) develop well-tested and standardized sampling, extraction, and isotope analysis protocols and (ii) verify whether these extraction methods faithfully return the water pools we actually aim to analyse.

Another issue that needs to be addressed is the pore-scale variation in soil-water isotopic composition. It has been suggested that differences in soil-water isotopic composition depend on the soil-water potential in different pore spaces, often referred to as “tightly bound” versus “mobile” water. Note that this language can be misleading because plants can use both. For example, plants sometimes rely on obtaining less-mobile matrix water that may have an isotopic composition distinct from the mobile gravity-drained water fraction that transits from the hillslope to groundwater and to streams (Brantley et al., 2017). However, soil and subsurface waters of different mobility were also found to be isotopically similar, and isotopic differences in these water sources can occur for reasons not related to mobility (McCutcheon et al., 2017). This suggests that our current perspective on why subsurface waters may vary isotopically is still limited. Currently, extraction of bulk water from soils or other subsurface compartments is done on relatively large soil sample volumes (50 cm3 and above; Sprenger et al., 2015). However, experimental designs and methods that target the isotopic composition of water that is bound with different potentials and distributed in different pore sizes on scales below 50 cm3 are needed to test whether such small-scale differences exist and to subsequently represent them in mixing models. If soil-water isotopic composition cannot explain the observed isotopic signature in plant (xylem) water, then other factors (e.g. plant physiological processes such as fractionation at the soil-root interface, uptake of dew, mist and/or fog via leaves and bark, evaporation through the bark, or mixing of xylem and phloem water) may explain these patterns (Eller et al., 2013; Berry et al., 2014; Martín-Gómez et al., 2016; Sprenger et al., 2017, 2018b; Lehmann et al., 2017; Dawson and Goldsmith, 2018).

Similar conceptual constraints relate to measuring the isotopic composition of water in plant tissue. It can take hours to days (or even months; Meinzer et al., 2006) for water absorbed by tree roots to reach the leaves (Dye et al., 1992; Ubierna et al., 2009). Further, water can be stored in the sapwood and outside the water transport pathway for days, particularly in conifers (Waring and Running, 1978; Meinzer et al., 2006). Thus, the isotopic composition of xylem water may not always reflect the current water source(s) used by plants but may instead be influenced by water taken up days or even months beforehand (Brandes et al., 2007; Treydte et al., 2014). Recent experimental and modelling studies have revealed that xylem isotopic signatures also vary on short, sub-daily timescales (Volkmann et al., 2016a), but that water taken up by plants may be a mixture of both young water (from the current growing season) and old water (precipitation from the previous year), also depending on the time of the year (Brinkmann et al., 2018; Sprenger et al., 2018a). So far, most studies have assumed – not explicitly but often implicitly – some kind of steady-state condition when trying to determine tree water uptake patterns. In addition, considerable spatial variation in xylem isotopic signatures within trees has been observed with values differing around and along tree stems (Cernusak et al., 2005; Volkmann et al., 2016a, b) and between stem and branch water (Dawson and Ehleringer, 1993; Cernusak et al., 2005; Ellsworth and Williams, 2007; Zhao et al., 2016). How these temporal and spatial variations in plant isotopic signatures can inform end-member determination, and how they can be integrated into mixing models, remain unsolved issues at present.

The “two water worlds” hypothesis (McDonnell, 2014) has challenged the assumption of complete subsurface mixing that underlies many catchment models (Pfister and Kirchner, 2017). This hypothesis postulates that more mobile soil water contributes to groundwater recharge and streamflow, whereas “tightly bound” water tends to be used by plants (McDonnell, 2014; Evaristo et al., 2015). Preliminary evidence from catchment studies based on the dual-isotope approach showed that bulk soil water was isotopically different from tension lysimeter water collected at the same depth (Brooks et al., 2010); that shallow soil-water pool utilized by plants differed in isotopic composition from precipitation, stream baseflow, and soil-lysimeter water pools (Goldsmith et al., 2012); and that xylem water may be isotopically similar to soil water and rainwater, but different from streamflow and groundwater (Penna et al., 2013). The two water worlds hypothesis has stimulated new interpretations of ecohydrological data and new research questions to investigate water flow pathways in catchments (McDonnell, 2014) but also calls into question how often such dichotomous conditions exist in natural systems (Brantley et al., 2017). Currently, there seems to be a tendency to focus on just confirming or rejecting this one hypothesis, although, as outlined in Berry et al. (2017) and Sprenger et al. (2016), alternative hypotheses need to be developed and tested to improve our current understanding. Because water held in the root zone or moving through that zone and other subsurface layers is a continuum, where water transport is driven by gradients, and not separate worlds, we see the necessity to move from the simplistic two water worlds hypothesis to an “n water worlds” concept, where multiple water reservoirs and flow pathways are invoked and parameterized, doing justice to the distribution of the different substrate types and sites. A challenge is that we are currently lacking easily applicable methods to sample the isotopic composition along this continuum. This can limit (and bias) sampling to only highly mobile and bulk soil water sampled by either tension lysimeters or cryogenic extraction and direct equilibration, respectively. Very recent findings highlighted that the relative contributions of mobile and less-mobile (bound at a wide range of water tensions) water are temporally variable and that the mobile water does not reflect the total plant-available water (Berry et al., 2017; Sprenger et al., 2018b). This should not be surprising: if water is highly mobile then it passes by roots too quickly for them to use. Although efforts have been made to compare different methods to sample water of different mobility for isotopic analysis (e.g. Geris et al., 2015; Orlowski et al., 2016b), we still lack clear definitions for distinguishing these multiple water pools. In addition to these methodological constraints, we also need to enhance our understanding of what pools of water plants might use and why. Some suggest that plants will use more “tightly bound” water even when more easily accessible, mobile water is available. As discussed by Bowling et al. (2017), plant-water uptake and transport within the plant are primarily physical processes driven by a potential gradient between soil and leaf. Thus, this notion of plants using “tightly bound” water is inconsistent with the well-established mechanisms of water uptake and transport in plants, i.e. via water potential gradients where plants are known to take up whatever water is most easily accessed (highest water potential) if they in fact have functional roots there (Dixon and Joly, 1896). The assumption that plants take up “tightly bound” water (as indicated by isotopic evidence) in the presence of less “tightly bound” water near the roots violates current physiological understanding about the mechanism of water uptake. Likely, other mechanisms, still not fully appreciated, affect the isotopic composition of water in plants. Therefore, we should inquire into them instead of invoking a notion about plant-water uptake that is inconsistent with previous investigations on plant physiology.

Given that the natural environment is heterogeneous on all scales, that sampling is by definition incomplete, and that the analytical process adds random errors (and, often, systematic biases), isotopic data are inherently subject to uncertainties. While many of our research questions are specifically focussed on exploring heterogeneous patterns across different domains (e.g. different climatological conditions, soil types or vegetation types), there is also considerable variability within each domain. Observed differences in the isotopic composition of water in various compartments of a catchment or an ecosystem are the result of many, often simultaneous, processes. The isotopic compositions of water samples are often shaped by mixing as well as fractionation processes; both mixing and fractionation can occur in different compartments (e.g. soils, plants, atmosphere), either simultaneously or at different times as water passes through the system. Because we still lack a thorough understanding of the underlying mixing and fractionation processes and of the spatial and temporal scales at which they operate, difficulties remain in interpreting the isotopic compositions that we measure in our environmental samples. While early isotope applications in catchment studies demonstrated the importance of considering temporal variability in precipitation and runoff (Kendall and McDonnell, 1993), less attention was initially paid to spatial heterogeneity, especially at the hillslope or small catchment scale, under the assumption that streamflow inherently integrates over spatial variations in the upslope contributing area. However, it is now well known that characterizing the spatial patterns in isotopic composition (so-called “isoscapes”) is another important tool for increasing our understanding of hydrological processes (West et al., 2009; Bowen and Good, 2015). In addition, applications in ecohydrology often require spatially explicit characterizations of soil water, groundwater, and plant water, which do not necessarily integrate across large spatial areas and thus reflect local heterogeneity. In the following, we highlight the main sources of heterogeneity that characterize different water pools relevant to ecohydrological studies.

Precipitation represents a major source of spatio-temporal heterogeneity that results in variations in all subsequent biological and hydrological compartments. The combined effects of variability in atmospheric parameters such as humidity, temperature, and solar radiation influence the isotopic composition of precipitation, manifesting in temporal (Dansgaard, 1964; Rozanski et al., 2013; Coplen et al., 2008, 2015; Munksgaard et al., 2012) and spatial (Ingraham, 1998; Bowen and Revenaugh, 2003; Bowen, 2008; Fischer et al., 2017; Allen et al., 2018) variability at multiple scales. However, at least on the plot scale (i.e. tens of metres) and in the absence of significant altitude variations, the spatial variability of precipitation is usually of minor importance. Precipitation is often collected by tipping buckets with a ∼200 mm diameter (Fig. 1), which is assumed to integrate small-scale variations and be representative of the plot scale. On larger scales (e.g. > hundreds of metres), we might observe systematic variations that are functionally relevant, and obtaining the isotopic composition of an input that is representative of the entire study area can be challenging. Temporal variability can also be very pronounced, even during a single storm event, and quasi-continuous precipitation sampling is vital for capturing such a variability in the input signal to the system (Munksgaard et al., 2012; Pangle et al., 2013; von Freyberg et al., 2017).

Canopy interception of liquid water involves flow through a small storage with short mean residence time, largely resulting in throughfall and stemflow having patterns of temporal variability that resemble the initial precipitation inputs (e.g. Ikawa et al., 2011). Nevertheless, storage and subsequent evaporation of intercepted precipitation can result in below-canopy inputs to soil that differ from open precipitation by more than 2 ‰ in 18O for single events and over longer periods (Allen et al., 2017). There can also be spatial variations of several ‰ in 18O because stemflow and throughfall dripping points involve long residence times (Allen et al., 2014) and thus have distinct isotopic compositions which are challenging to capture.

In places where snowfall is an important component of precipitation, snow accumulation dynamics can significantly modify the spatio-temporal patterns of precipitation isotopic composition. Snowpack depth and density are known to be very irregular, following complex compaction and redistribution dynamics that are influenced by topography, wind and vegetation (e.g. Trujillo et al., 2009). As a consequence, snowmelt is very heterogeneous and its flow pathways change through time as the snowpack evolves. Due to these dynamics, the spatial variability in the isotopic composition of snowpacks and snowmelt can be very large (Rücker et al., 2019; Webb et al., 2018). Moreover, melt-and-refreeze dynamics during water percolation through the snowpack cause heterogeneous and time-variable isotopic fractionation (Taylor et al., 2001). Canopy-intercepted snow can have longer residence time than liquid water and, because fractionation due to sublimation and refreezing is greater, especially at lower temperatures, it may contribute to larger isotopic changes (Koeniger et al., 2008). All these processes often interact and make the representative characterization of snow and especially snowmelt isotopic composition highly difficult.

Water flow and transport through heterogeneous porous media are complex processes that still represent a hot topic for the vadose zone and groundwater communities (Kitanidis, 2015). However, not only is subsurface flow always heterogeneous (Gehrels et al., 1998; McDonnell et al., 2007; Stumpp et al., 2007; Troch et al., 2009; Stumpp and Maloszewski, 2010), but fractionation processes in the subsurface also vary in both time and space. Evaporation, which is largely controlled by surface energy variations, is a major contributor to isotopic fractionation, especially at shallow soil depths. Generally, evaporative effects decrease with increasing soil depth, resulting in deuterium excess or line-conditioned excess of soil water becoming less negative with depth (e.g. Sprenger et al., 2017). Moreover, bulk soil water shows more evaporative effects than lysimeter water and may be characterized by values of line-conditioned excess below zero (McCutcheon et al., 2017), because more “tightly bound” waters integrate older ages and, therefore, are affected by kinetic fractionation during periods of atmospheric evaporative demand (Sprenger et al., 2017). However, bulk soil water, extracted by cryogenic distillation or direct equilibration (Sprenger et al., 2018c), which contains both mobile and matric-bound water, is generally more depleted in heavy isotopes than mobile water collected by tension lysimeters at the same depth and location. It is unclear how much, volumetrically, the “bound water” fraction in the bulk sample is, what isotopic impact it has on the final measured isotope ratio, and how much of it a plant might or can use (Brooks et al., 2010). Moreover, deep bulk water is usually more depleted than mobile water during spring or summer due to filling of fine pores of a relatively dry soil with depleted precipitation several months earlier (Geris et al., 2015; Oerter and Bowen, 2017; Sprenger et al., 2017). This suggests that old and more “tightly bound” water might show not only a distinct isotopic signal compared to mobile water due to seasonally variable precipitation inputs, but also an evaporative enrichment signal from periods of high evaporative demand (Sprenger et al., 2017).

In soils and groundwater, isotopic heterogeneity results from differences in the inputs (precipitation, throughfall, snowmelt), differences in the temporal integration of previous precipitation events (Yang et al., 2016), and differences in the subsequent fractionation from evaporation and transpiration (Benettin et al., 2018). As a general rule, smaller storage-to-output ratios with short residence times generally lead to higher temporal variability. Conversely, larger storages with longer residence times are likely to lead to a more dampened signal that integrates over longer periods of time (e.g. Zhang et al., 2016; Benettin et al., 2017). Obtaining representative samples in soils is also challenging because soil water content, soil texture, mineral composition, and the content of organic matter are spatially heterogeneous and strongly influence how soils interact with water molecules (e.g. Barnes and Allison, 1983; Oerter et al., 2014; Oshun et al., 2016; Gaj et al., 2017a). Hence, the interaction with soil particle surfaces (Lin et al., 2018), soil organic matter (Orlowski et al., 2016a), local soil properties (Yang et al., 2016), microorganisms (Blake et al., 1997; Kool et al., 2007), and plants (Vargas et al., 2017) may introduce additional isotopic heterogeneity. Also, the origin of soil water is diverse, comprising a mixture of precipitation events from different times, sources (air masses), and types (e.g. rainfall, snow, hail); it may also include groundwater and, in the case of agricultural fields, irrigation water derived from groundwater, lakes, or rivers. With increasing soil depth and down to groundwater, we generally expect that dispersive transport will lead to increasingly damped spatio-temporal variations around the average input composition. Nevertheless, we commonly characterize groundwater with wells that receive water from a variety of depths that may have isotopically distinct waters (Jasechko et al., 2017). While it is generally assumed that groundwater integrates inputs over time and space (Scheliga et al., 2017), the integrations may be short and small when one distinguishes shallow groundwater or perched water tables from soil water (Uhlenbrook and Hoeg, 2003). Smaller-scale spatial variations in groundwater isotopic composition are typically not well characterized, and ecohydrological applications often assume spatial homogeneity.

Plant xylem water reflects the spatial and temporal heterogeneity of the water that is accessed by functional plant roots (Ehleringer and Dawson, 1992). Due to differences in functional rooting locations among species and individuals, plant-water isotopic composition is often distinctly different and highly variable among different species (Bertrand et al., 2014; Schwendenmann et al., 2015; Volkmann et al., 2016a). However, heterogeneity in soils and other subsurface compartments with respect to texture, structure, and water content could also result in differences in xylem water isotopic composition across trees with similar rooting patterns. Within-plant variations in xylem water composition may also occur because travel times can increase with within-plant path length, so temporal variations in source isotopic composition must result in vertical variations in xylem isotopic signatures (Dawson and Ehleringer, 1993; Cernusak et al., 2005; Zhao et al., 2016). Lateral or radial variations in the trunk occur due to radial variations in the source water composition or potentially in water transport rates (and thus water age), depending on the degree of sectoriality of a plant's xylem transport (Steppe et al., 2015; Volkmann et al., 2016b). Even among individual vessels, variations are expected, given the little evidence of dispersion across xylem flow pathways provided by different studies (Zimmerman and Brown, 1971; Kline et al., 1976). Once water in the xylem reaches the stomatal aperture of the leaves, the vapour pressure deficit between the ambient air and the intercellular cavities results in isotopic fractionation that significantly alters the source signal (Dongmann et al., 1974). These fractionation effects are spatio-temporally variable (see, for instance, Helliker and Ehleringer, 2002) but may be masked by wood and other tissues that might act as temporal and spatial integrators of heterogeneous processes in leaves (Gessler et al., 2014; Singer et al., 2014).

Many of the issues raised in the above sections relate to small-scale processes of water flow and transport, as research to date has often been performed at the level of soil patches or individual plants. Less attention has been directed towards determining how small-scale ecohydrological processes can be used to understand catchment- or landscape-scale phenomena. For example, Bertrand et al. (2014) found that trees used different water sources depending on their location within an alluvial system, whereas Gaines et al. (2016) did not detect differences in root length density and water uptake among trees located at different hillslope positions. Hsueh et al. (2016) showed that trees on deltaic hummocks preferentially took up water from unsaturated hummock tops to protect from higher salinity and saturated soil in swales and the lower portions of hummocks. Dudley et al. (2018) saw little evidence to suggest that landscape position altered groundwater uptake by shrubs. Pettit and Froend (2018) showed that riparian trees located on relatively shallow groundwater had greater growth rates, larger diel responses in stem diameter and were less reactive to extended dry periods than trees located in areas of deep groundwater. Sprenger et al. (2018c) found that soil water beneath conifer trees was more fractionated than beneath heather shrubs or red oak trees and that sampling locations closer to streams had a more depleted isotopic composition than hillslope sites, revealing increased subsurface mixing towards the saturated zone and a preferential recharge of winter precipitation. Such differences in the depths and types of water taken up by trees may have a critical impact on streamflow and hydrochemistry (Brantley et al., 2017), both of which generally depend on travel times and flow pathways (Rinaldo et al., 2015; van der Velde, 2015). Studies that systematically monitor ecohydrological processes across environmental conditions, soils, and vegetation types within landscapes will certainly be needed in the future, but a challenging question today is how representative such individual tree/plant studies are for larger-scale systems. Answering this question will require us to understand both the heterogeneity in ecohydrological processes and the resulting heterogeneity in the isotopic composition of water when moving up in scale. As we move, for example, from the individual to a stand or hillslope scale, systematic sampling approaches that account for this heterogeneity within a landscape element of interest will have to be tested. These approaches might identify a sample size that is statistically representative and integrates over the main sources of variation within a given system.

If such an approach exists for a given process of interest, it could inform sampling protocols. As we move further up in scale, e.g. to catchments or an entire landscape, we need to develop approaches that appropriately represent this heterogeneity in hydrological and ecological models. Here, much may be learned from work that has been carried out in soil science (Lark, 2012a, b) on how to investigate spatial patterns and scaling related to isotopic studies in ecohydrological systems.

Furthermore, a gap exists between the scale at which we typically apply our observational isotope techniques and the range of spatial and temporal scales across which we draw ecohydrological inferences. Repeated observations have provided insights into the heterogeneity of hydrological and ecological processes on these larger scales, but they have also revealed the limitations in our current sampling strategies. In defining the isotopic signal of the measured ecohydrological compartments as their variations across spatial and temporal scales, and the noise as the variance of repeated sampling of the same water pool, the key issue is to determine which sampling strategies can ensure the best signal-to-noise ratio, i.e. that allow for the signal to be many times greater than the noise. However, inevitably, our ability to observe the variability within a sampling event is a product of the duration and size of that sampling. In practice, we often rely on a few samples to characterize a much larger heterogeneous domain. For example, small samples of xylem tissue are commonly used to characterize an entire forest or a few soil cores are meant to represent the entire range of spatial variation in soils. Temporally variable processes are often sampled in a few short measurement campaigns, yet they are assumed to be representative of much longer (and perhaps variable) periods. Furthermore, investigators should consider whether concurrently sampled components of the hydrologic cycle are representative of the same time (e.g. leaf water may be sourced from soil water taken up by roots days or weeks prior to the day of sampling). Inadequately sampling heterogeneous domains can lead to interpretation errors. This problem is exacerbated when analysis methods do not properly reflect the statistical properties or uncertainty of the sample pool. For example, single mean values are often used as end-members in mixing models, which exaggerates the precision of the source partitioning (Phillips and Gregg, 2001).

We recommend that researchers adopt sampling strategies to determine the signal variances and to deliberately integrate across the smaller-scale variability so that heterogeneity across the ecohydrological domains of interest can be appropriately characterized. This includes, for instance, planning sampling campaigns that span over multiple (growing) seasons or that comprise multiple locations within the same landscape element (e.g. different hillslope positions), finding the correct balance between the necessary labour and cost, and the additional information provided. For example, this would allow for considering mean values with their variability as end-members in mixing models, thus more appropriately addressing the uncertainties arising from the intrinsic spatio-temporal variability in the studied system.