Discussion

Xylem water matches source water, but bulk stem water doesn’t

Our data show that the Cavitron-extracted xylem water had δD signatures that closely aligned with source water (mean of δD offsets not significantly different from zero), with virtually identical trend lines between these two groups. This result corroborates the findings of recent studies that used a Cavitron for xylem water extraction (Barbeta et al., 2022; He et al., 2023; Wen et al., 2023). Similar centrifugation methods relying on smaller stem segments have also resulted in relatively good alignments with source water (Sánchez-Murillo et al., 2023). While not directly tested against source water measurements, other techniques targeting mobile xylem water, such as direct vapour equilibration (Millar et al., 2018) and pressure chamber extraction (Bowers & Williams, 2022; Zuecco et al., 2022), have also produced δD signatures likely reflective of source water.
We found that unlike xylem water, bulk CVD-extracted water was strongly depleted in δD relative to both source water (median δD offset –15.5\($\textperthousand$\)) and Cavitron-extracted water (median δD bias –14.9\($\textperthousand$\)). The tendency of CVD extraction to introduce systematic isotopic biases is now well-known and has been thoroughly described through recent experimental work by Wen et al. (2022). These authors have shown that CVD-induced bias is caused by both within-stem isotopic differences between xylem and tissue (as per Barbeta et al., 2022) and hydrogen exchange with organics (as per Chen et al., 2020), with hydrogen exchange being the dominant process. Diao et al. (2022) propose an alternative perspective, suggesting that while hydrogen exchange does occur, CVD-induced biases are mostly related to isotopic fractionation during CVD extraction, and that extracting larger volumes of plant water through CVD can reduce these biases. In our case, the exact CVD-extracted volumes were not recorded, but were on average 0.25 mL. According to Diao et al. (2022), this small sample volume size could lead to a δD bias roughly ranging from –10 to –35\($\textperthousand$\), consistent with our observations – although we note that the volume effect would be specific to the CVD setup used.
Overall, our results add to the growing empirical evidence that sampling xylem water yields more reliable estimates of plant water sources, compared to methods analysing bulk stem water.

Cryogenic bias is species-specific, but independent of stem water content and status

Our data indicate that while the CVD-induced δD bias was apparent across seven tree species, it affected each species differently, with mean biases ranging from –19.3\($\textperthousand$\) (C. bella) to –9.1\($\textperthousand$\) (M. argentea). These results support the findings of Chen et al. (2020), who showed via rehydration experiments that CVD-induced biases differed in the range –5 to –11\($\textperthousand$\) across nine species, and those of Barbeta et al. (2022) who observed significant inter-specific differences in the range –12.7 to –22.3\($\textperthousand$\) across three species. More broadly, the global synthesis by de la Casa et al. (2022) suggests considerable variations in δD biases across different species – although this study inferred biases based on δD offset calculations using source water, rather than through direct measurements.
Variability in δD biases within single species was relatively low, although the small number of replicate trees was a limitation of our study – except for H. arborescens for which we have data for seven individual trees. For this species, the measured δD biases varied between –20.4 and –13.6\($\textperthousand$\). Limited within-species variation was also reported by Barbeta et al. (2022) for Fagus sylvatica and by Bowers and Williams (2022) for a range of conifer species. Wen et al. (2022) found large variations in the δD of xylem water between apple trees of the same species (and even within single individuals), but their δD biases were less variable, similar to our findings.
CVD-induced δD biases were not correlated with RSWC or stem water isotopic composition. This is at odds with recent laboratory studies where stem water content emerged as a key driver of the CVD δD bias. Chen et al. (2020) and Wen et al. (2022) showed that a higher stem content resulted in a lower bias, and Wen et al. (2022) showed that more δD-depleted stem water resulted in lower biases. However, these two studies were rehydration experiments so may not be reflective of natural conditions. Our results are more in line with those of Barbeta et al. (2022), Bowers and Williams (2022) and He et al. (2023) who found, under field conditions, non-significant or weakly significant relationships between RSWC and δD bias.
CVD-induced δD biases were not correlated with pre-dawn LWP either. This lack of a relationship between δD biases and water availability suggests that the extent of tree water stress may not affect the CVD bias. Bowers and Williams (2022) observed a negative correlation between species-specific xylem vulnerability to cavitation and δD bias, and hypothesised that less vulnerable species might have less well-mixed xylem conduits, potentially leading to higher δD biases. While our data do not allow us to test this hypothesis, it is plausible that the observed inter-specific differences may be related to anatomical differences between species, rather than to point-in-time water stress conditions (e.g. differences in connectivity of xylem conduits, variable xylem residence times; Bowers and Williams (2022)). The wide range of observed LWPs (–1.97 to –0.18 MPa) suggests that the sampled species may represent a spectrum of anatomical adaptations to aridity, given this is a strong driver shaping stem hydraulic traits in Australian trees (Peters et al., 2021). Overall, there is a clear need for research that further untangle the respective roles of anatomical and functional tree properties in CVD-induced δD biases.

Concluding remarks

Our dataset provides robust evidence of (1) a strong δD bias in CVD-extracted bulk stem water relative to xylem water and source water, (2) significant differences in the magnitude of these biases among tree species, and (3) the limited influence of RSWC and LWP in explaining variations in δD bias. However, our inability to extract sufficient water from some stem samples resulted in a low number of replicates per species. These low numbers might have hindered the detection of any species-specific patterns in the data, suggesting that our third conclusion might not hold for individual species. To improve water extraction yields, particularly for trees in seasonally dry environments, we recommend using a larger version of the Cavitron (500-mm diameter), which can host longer stems hence yield higher water volumes.
In the context of future plant water sourcing studies, we recommend that non-destructive extraction techniques that target xylem water, such as Cavitron centrifugation (e.g. Barbeta et al., 2022; He et al., 2023), pressure chamber (e.g. Wen et al., 2023; Zuecco et al., 2022) or in-situ techniques (e.g. Kübert et al., 2023; Kühnhammer et al., 2022), be preferred over CVD extraction. Should no alternative be available, one should ensure that large volumes (specific to each experimental setup) are extracted via CVD (Diao et al., 2022). In any case, the similarities we found between average δD biases and average δD offsets lead us to the conclusion that CVD-derived bulk stem water isotopic signatures can potentially be corrected to provide a reasonable approximation of xylem water isotopic signatures. Yet this adjustment can only be done using a site-specific δD offset, a step that requires the local source water line to be known.
Unlike Chen et al. (2020), we discourage the indiscriminate use of a uniform offset correction for all sites, because δD offsets may be highly site- and method-dependent (Diao et al., 2022; Millar et al., 2018). We also recommend using an average, site-specific δD offset for the correction of CVD data (Duvert et al., 2022) rather than individual offsets (Barbeta et al., 2019; He et al., 2023), as using individual offsets for each sample eliminates the natural variability in δD among samples. In turn, this can introduce additional uncertainties to plant water source identification. Until a complete understanding of the mechanisms generating δD offsets is achieved, corrections of CVD data should be made with a high degree of caution, and researchers should consider assessing plant water sources based on δ18O data alone.
Our work emphasises how considering an appropriate methodology when seeking to characterise plant water uptake is key to advancing our understanding of the role of vegetation in partitioning rainfall into evapotranspiration and recharge. From a water resource management perspective, this field of research is also becoming increasingly relevant, particularly in areas where exploited groundwater systems support groundwater-dependent ecosystems of ecological and cultural significance.