This study was conducted in a ∼100-year-old naturally regenerated, homogeneous forest dominated by Scots pine (Pinus sylvestris L.), in northern Sweden (Rosinedalsheden, 64°10^′ N, 19°45^′ E, 145 ma.s.l.). Understory vegetation was composed of ericaceous shrubs (Vaccinium myrtillus L. and Vaccinium vitis-idaea L.) and a layer of mosses and lichens. In 2013, the leaf area index was 2.7 m2m-2. The stand averaged 18.6 (SD =0.1) cm in diameter and 17.5 (SD <0.1 m) m in total height in 2013 (Lim et al., 2015). The climate is typical of the northern boreal zone: average annual temperature and precipitation are 2.4 (SD =0.8) °C and 638 (SD =107) mm, respectively (Klosterhalfen et al., 2023). Winters are cold and long; summers are usually cool and wet. The photosynthetically active period extends from mid-April to mid-November (Tarvainen et al., 2018; Vernay et al., 2020). A soil description was performed at a nearby forest, approximately 6 km away, under similar stand and landscape conditions. The soil was labeled a Regosol that had formed on fluvial sediment. The texture was considered loamy sand to sandy loam with lenses of silt loam at 25–45 cm depth. The clay content was less than 11 % (Plamboeck et al., 1999).

The sample plot was located approximately 30 m from an eddy covariance system installed on a tower above the uniform canopy (21.5 m above the ground) (Lindroth et al., 2020; Zhao et al., 2022). The eddy covariance system measured latent energy fluxes from the canopy over the whole study period using H2O mixing ratios measured with a gas analyzer (LI-7200, LI-COR Environmental, Lincoln, USA) at 20 Hz frequency (Jocher et al., 2017).

The study was conducted from June to September 2018, when northern Scandinavia, like much of central Europe, underwent an exceptional drought; in Scandinavia, it was the driest period in fifty years (Gutierrez Lopez et al., 2021; Lindroth et al., 2020). From snowmelt (18 May) until 23 July, a period of more than two months, the study area received only 56 mm precipitation. The drought was then broken by a series of strong rainfall events that delivered 85 mm in eight days. We applied the soil label in the middle of the drought period and tracked its passage through the last two weeks of the drought and into the wetter period afterward.

Volumetric soil water content (VWC) was measured along a vertical profile inside the labeled area using TDR probes (5TM, Metergroup, Pullman, USA). The TDR probes were placed at 10, 30, 50, 70, and 100 cm soil depth. For our sandy soils, we used the factory-default calibration. We took particular care when installing both the TDR and the isotope probes (see Sect. 2.3.2) to avoid damaging roots during probe installation, hoping that the roots from all surrounding trees would continue to take up water normally from the labeled block of soil. We estimated total water uptake by measuring the daily decline in VWC at each depth, multiplying VWC by the soil depth represented by that sample, and accumulating the decline over all depths (the single-step, multi-layer water balance of Guderle and Hildebrandt, 2015): 1ΣU=ΣΔVWC10⋅D10+ΔVWC30⋅D30+ΔVWC50⋅D50 Where ΣU is the sum of uptake per day, ΔVWCn is the daily change in VWC at depth n, Dn is the thickness of the layer represented by each VWC (always 20 cm). This approach yields a water consumption rate (ΣU) in cmd-1, integrated over the upper 60 cm of mineral soil. Because the uppermost TDR probe was located 10 cm deep in the mineral soil and was covered by the forest floor, we assumed there would be little contribution of soil surface evaporation to this measured change in VWC and that it would therefore be dominated by root uptake. We return to the broader topic of surface evaporation in the discussion.

We then measured the mean depth of uptake was by weighting ΔVWCn⋅Dn of each layer by its mean depth, summing, and dividing by the total uptake, thus: 2UD=ΣΔVWC10⋅D10⋅10+ΔVWC30⋅D30⋅30+ΔVWC50⋅D50⋅50/ΣU Where UD is mean uptake depth and 10, 30, 50 cm are the depths of each VWC sensor.

The experiment was designed primarily to detect and quantify patterns of root (and mycorrhizal) overlap and dynamics of horizontal water uptake by trees. We found the label in stems of six surrounding trees, meaning that the root systems of individual trees do not express complete territoriality at this scale. We therefore conclude that all these trees had roots in the labeled 1 m2 plot. This is far more overlap than in the crowns, which scarcely overlapped at all.

There are relatively few studies describing horizontal water uptake patterns in forest trees (Göttlicher et al., 2008; Henriksson et al., 2021; Lutter et al., 2021). Root distributions have been mapped in some temperate forests (Gao et al., 2021; Kuiper and Coutts, 1992; Lwila et al., 2024), but the presence of roots is not sufficient to predict water uptake (Göttlicher et al., 2008). Long-distance horizontal transport through overlapping root systems might increase uniformity in water availability among trees. It might also favour trees near openings by providing access to the unused soil water in the openings (Lutter et al., 2021; Moreira et al., 2000; Sternberg et al., 2002, 2005). This favouring might, as with deep vertical roots, provide a “lifeline” to trees growing under extreme drought conditions (Bachofen et al., 2024). However, it might also represent a significant carbon cost (Guswa, 2008), perhaps reducing partitioning to aboveground production (Fernandez-Tschieder et al., 2024; Marshall et al., 2023).

The rate of uptake varied among tree individuals. For example, the stem of Tree 2 was at the edge of the labeled plot (0.7 m from the center of the plot) and yet it showed far less label than Tree 1, which was slightly farther away (Fig. 5a and b). The label intensity detected in tree xylem became rather weak with greater distance from the plot, which we addressed by calculating the lc-excess. This parameter corrected for event-based variation using the strong correlation between natural abundance of δ2H and δ18O, allowing us to detect low levels of labeling. When we focused simply on the first appearance of the label, as defined by lc-excess greater than zero, we found that there was a correlation with distance from the plot (Fig. 6), suggesting that the label took some time to traverse the horizontal distance to the stems. The slope of this relationship, 0.17 (0.05) md-1, is an estimate of the maximum horizontal transport rate. For comparison, measurements of water flow through pine stems are typically around 1 md-1 (Plamboeck et al., 1999; Tarvainen et al., 2018). Note also that the label appeared in Tree 1 in only four days. This suggests that the surface roots were alive and ready to begin water uptake, despite the severe drought (Marshall, 1986). The peak label intensity in Tree 1 appeared eight days after the labeling, perhaps due to the arrival of label via slower flow paths after its initial breakthrough (Gao et al., 2021; Seeger and Weiler, 2021; Werner et al., 2021). Alternatively, as the drought advanced, the tree could have increased the amount of water taken up from deeper soil layers (e.g., Gessler et al., 2022); however, our VWC data suggest that deep water uptake did not increase as a proportion of the total (Fig. 3c) late in the drought.

We also observed a brief, weak pulse in δ18O of the surface soil horizons after the labeling. This might have occurred due to δ18O-enrichment of the 2H2O used for labeling (not measured), but we believe it is more likely interference in the spectrometer by the strong 2H1H16O absorption peak with the adjacent 1H1H18O peak, which has been observed in other studies using high concentrations of δ2H (Beyer et al., 2016, 2018). In any case, the apparent δ18O pulse was orders of magnitude weaker than the δ2H pulse and can be considered negligible.

The water uptake benefits of long horizontal roots could be assessed not only by the detection of label in surrounding trees, but also by the label intensity. One can imagine a continuum between complete domination of uptake by a single tree, analogous to owned and defended property, vs. complete sharing of the benefit by all trees within some radial distance from the tree stem. The distinction is important, in part, because conservation of soil water for later use would be favored by territorial rooting and disfavored by extensive root overlap (Cohen, 1970). Although six trees were labeled in our plot, we detected a strong preference for uptake by one of the two trees nearest the plot center, while the other tree at a similar distance took up very little of the label. This preference remained remarkably constant even as the label pulse dissipated (Fig. S2).

In order to investigate this territoriality further, we have used the data from this earlier experiment to rank the trees from each of their plots according to label intensity (Henriksson et al., 2021) (Fig. 7). The ranking shows a strong preference for the labeling of one or two trees, with much lighter labeling in several more (Fig. 7b). Henriksson et al. (2021) found that the most heavily labeled trees were not necessarily those nearest the labeled plot. We interpret these and our results as evidence of territorial water uptake, but not to the point of exclusion.

Our results may have been influenced by choosing a microsite with particularly high tree density (1600 trees per hectare relative to 1010 (SD =125) trees per hectare in the stand, Lim et al., 2015). This was done to facilitate access to the central isotopic analyzer. However, in an earlier study in the same stand with 1 m2 water-labelled plots, Henriksson et al. also reported 8 (SE =2) labelled trees per plot. We therefore conclude that our results were not atypical.

The territoriality interpretation draws attention to the orientation of the trunk boreholes in our study. Ours were oriented north–south while the boreholes in Henriksson et al. (2021) were oriented toward the plot center. We initially believed that borehole orientation would not matter because an earlier report had shown that label was evenly distributed throughout the tree circumference rather quickly in white pine (White et al., 1985). More recent evidence from our site (Tarvainen et al., 2018) suggests that the stem label distribution is not uniform near the ground. Our borehole orientation (N–S) may have contributed to the weak labeling observed for Tree 2 (on the western edge of the labeled plot) compared to Tree 1 (to the northeast) despite the proximity of Tree 2 to the labelled plot. Therefore, we recommend orienting boreholes toward the center of small labelled plots in future work.

We interpret daily declines in soil VWC during the drought period as evidence of tree root water uptake throughout the upper 50 cm of mineral soil (Fig. 3a). The slopes of the declines at 10 and 50 cm are similar, but that at 30 cm is steeper. There was no evidence of change in water content at 70 cm, consistent with our informal observation that roots were rare below 50 cm depth. The integrated water loss rate averaged 0.057 cmd-1 in the early part of the drought and increased to 0.083 cmd-1 in the last part, meaning that water uptake from these depths increased slightly as the drought proceeded. However, this conclusion is based on only three measurements across depth, the uppermost at 10 cm. Note that if there were evaporation in addition to root water uptake from the 10 cm depth, it would be included in water uptake depth estimate. Although there was almost certainly some evaporation from the injection sites at the surface of the mineral soil (0 cm depth), we speculate that evaporation was small at the measured 10 cm depth.

The maximum amount of label found in the soil was 4 %–5 %, which is small in comparison to the amount added (see “Partial Label Recovery” in the Supplement). However, we did not measure evaporation from the soil surface during the labelling and thereafter, which probably consumed a majority of the remainder. We tried to minimize surface evaporation by injecting at the mineral soil surface, but perhaps the injected water was drawn back up into the litter and moss and humus layers and above the topmost soil probes by matric forces during the injection.

After the injection, the labelled water was expected to flow downward under unsaturated conditions for as long as the water potential exceeded field capacity (Bonan, 2015). Remembering that the drought had already begun when the label was applied, the downward flow was expected to cease when field capacity was reached. One might expect a bit of short-range mixing by diffusion to continue after this point. This is consistent with the timing and intensity of the label peaks in Fig. 4a and b, where the label intensity decreased sharply between 15 cm (2027 ‰ maximum) and 20 cm depth (400 ‰ maximum, 8 d later). The added label could also have escaped detection by flowing down preferential flow paths (Clothier et al., 2008; Weiler, 2017), e.g. due to differences in soil density or along macropores or tree roots, leading to uneven infiltration and potentially bypassing probes and sensors. This might explain, for example, the higher maximum labelling at 70 cm than at 50 cm and at 30 cm than at 20 cm. However, the labelling patterns are largely consistent with the field-capacity scenario.

The current study follows on the classic study of Plamboeck et al. (1999), which was conducted at a nearby site with similar soils. Plamboeck et al. (1999) estimated that, without irrigation, the 2 cm humus layer provided 10 % of total uptake in July. When we estimated the mean depth of water uptake (Fig. 3c), it was remarkably stable throughout this severe drought, with a slight upward shift toward the end. Our estimated mean uptake depth (30.6 (SE =1.2) cm) was similar to a depth-weighted estimate based on Plamboeck et al.'s (1999) isotopic data (30.3 (0.07) cm). Both our results and those of Plamboeck et al. (1999) suggest a rather constant water uptake depth as a drought develops (Fig. 3c). This differs from a global meta-analysis, which found a downward shift in response to drought (Bachofen et al., 2024). If our trees cannot shift deeper under drought, then it may make the broad horizontal root distribution even more important as an emergency source of soil water.

The method used to measure xylem water here included several simplifications of the borehole method used in previous studies. In particular, we did not use dry air (e.g., Kühnhammer et al., 2022) relying instead on the two-way exchange and equilibration of water vapor during the borehole passage (Marshall et al., 2020). In addition, we relied on the pump within the analyzer to draw air through the boreholes under negative pressure, apparently without significant air leakage into the downstream flows. This eliminates the need for the upstream pump used in some other applications (e.g., Kühnhammer et al., 2022). Finally, we did not use any screening material to inhibit microbial growth (Landgraf et al., 2022), relying instead on the natural defensive compartmentation of the xylem to serve that purpose (Shigo, 1984). Although a visible band of brown tissue formed around the borehole, our data showed that the borehole vapor continued to respond to xylem-water changes until the end of the experiment, more than three months after the boreholes were drilled. This is somewhat surprising given that the drilling must cavitate the xylem conduits that are cut, but our data, like previously published results with this method (Kühnhammer et al., 2022, 2022; Marshall et al., 2020), suggest that hydraulic contact with the xylem flow is maintained. As a reminder, we note that modeled results suggest that the equilibration is so fast that it would best reflect the isotopic composition of the water in the last few mm of the borehole (Marshall et al., 2020). Fortunately, this is where the flow is fastest and the circumference is greatest, both contributing to high proportions of the total stem flux.

One complication that remained was the placement of an analyzer in the field. Several recent papers describe methods for field sampling and storage of water vapor for stable isotope analysis (Havranek et al., 2020; Herbstritt et al., 2023; Magh et al., 2022), which would mean sacrificing the continuous data streams collected here, but would circumvent the need for an analyzer in the field and provide opportunities for more replication.

These results provide a dynamic description of soil water storage, transport, and root water uptake, including passage of the label through surrounding tree stems as they take up the label for transpiration. The timing and detection of the label are consistent with expectations insofar as it supports the notion that the root systems of several trees overlap on each square meter of ground surface. However, the concentration of the label in a single tree was inconsistent with the notion of extensive horizontal sharing. This question deserves more attention as it influences the scaling of transpiration from trees to stands.