The study was conducted in 2019 in Chengcheng County, Shaanxi Province, China, in the temperate continental monsoon climate zone. Mean annual precipitation in the study region is 507.9 mm yr-1, and the annual average temperature is 12.6 ∘C yr-1 (1999–2018). A thick layer of loess covers the land surface, and the groundwater level is more than 50 m in depth on average, which cannot be absorbed by plant roots. Apple (Malus pumila Mill.) orchards of two stand ages (11- and 17-year-old) were selected (Fig. 1) to collect soil and xylem samples. The orchards are located in the same small watershed, with similar slopes, aspects, and soil texture, and subjected to the same management regimes (e.g., plant and row spacing (4×4 m), no irrigation, standard clipping and fertilization). The sampling locations, height, trunk diameter at 80 cm height, and crown dimensions (long and short axes) of trees were recorded. General information on the apple orchards is in Tables 1 and S1 in the Supplement.

A method combining isotopic labeling and the natural abundance of stable isotopes was used to investigate the water use strategy of apple trees. Firstly, isotopic labeling in deep layers was used to identify whether trees absorb DLSW and to determine the soil layer depths from which trees derive their water source. Next, we used the natural abundance of stable isotopes method to quantify the water use strategy of trees. Plant and soil samples in two experiments were collected at three developmental stages in 2019: blossom and young fruit (BYF, May), fruit swelling (FSW, July), and fruit maturation (FTM, September).

In the FTM stage, a hand auger with 60 mm internal diameter was used to collect root samples (50 cm from the tree trunk) and three trees in each orchard were selected randomly for sampling. The samples were collected down to 500 cm in increments of 20 cm. The samples with roots were washed carefully with tap water in a sieve (0.2 mm) to remove all the soil. The root samples obtained were scanned using a scanner at 300 dpi, and then the fine root length was determined using WinRhizo software (version 5.0 Regent Instruments Inc., Quebec, Canada). The fine root length density (FRLD) in each sample was calculated by dividing the length of fine roots by the volume of the sample.

A CVD system (Li-2000; LICA United Technology Limited, Beijing, China) was used to extract water under a heating temperature of 95 ∘C and a pressure of 0.2 Pa, which has been applied in previous studies (Huo et al., 2020; Tao et al., 2021a; Wang et al., 2021b; Zhao et al., 2020). The extraction time of soil water and xylem water samples were 90 and 120 min, respectively. Samples were weighed before and after extraction and again after oven-drying for 24 h to calculate the extraction efficiency (Wang et al., 2021b). Samples with an extraction efficiency less than 98 % were discarded. The stable hydrogen and oxygen isotope compositions of extracted soil water and xylem water were determined using an isotope ratio infrared spectroscopy analyzer (TIWA-45EP, Los Gatos Research, Mountain View, USA) and stable isotope ratio mass spectrometer (Isoprime Limited, UK), respectively. The measurement precision of δ18O and δD is 0.2 ‰ and 1.0 ‰ for the TIWA-45EP isotope ratio infrared spectroscopy analyzer and 0.3 ‰ and 2.0 ‰ for the stable isotope ratio mass spectrometer, respectively. Each isotopic sample was repeatedly measured six times. The first three measurements were discarded to mitigate the memory effect of isotopic measurement, and the mean value of the last three measurements was taken as the isotopic value of sample.

In general, soil water is the primary water source for trees on the Loess Plateau. We assessed the isotopic offset between xylem water and soil water using soil water line-conditioned excess (SW-excess) proposed by Barbeta et al. (2019): 1SW-excess=δD-asδ18O-bs, where as and bs are the slope and intercept of soil water line (SWL), respectively; δD and δ18O are the isotopic compositions of xylem water. A positive SW-excess value means that xylem water plots above SWL in a δD – δ18O diagram (i.e., D in xylem water is more enriched than SWL), while a negative value means that xylem water plots below SWL in a δD – δ18O diagram (i.e., D in xylem water is more depleted than SWL).

The Bayesian isotope mixing model, MixSIAR (version 3.1.7) package in R was used to calculate the contributions of soil water from different layers to xylem water (Stock and Semmens, 2013). Based on the distributions of soil water isotopic values and the results of labeling experiments, the 0–320 cm soil profile was divided into four water sources (0–40, 40–140, 140–240, and 240–320 cm layers). The SWC and isotopic values varied the most in the shallow-layer soil (0–40 cm) but were most stable in the deep-layer soil (140–320 cm). The δD and δ18O values for each potential water source were used as source data; the δD after subtracting the SW-excess and δ18O values for xylem water were used as mixture data for the model. δD values of xylem water corrected by SWL can match those of soil water. Thus, the fractionation factor was set to zero, assuming no isotopic fractionation during root water uptake.

A one-way analysis of variance was used to determine differences in SWC, δD values, and the contribution of water sources among sampling events. The least significant difference was used to perform the post-hoc analysis, with significance evaluated at the 0.05 level (P<0.05). The statistical analyses were carried out in SPSS 23.0, with all figures drawn using Origin 2016.

Figure 2 shows the total precipitation (Pt) and growing season (April to September) precipitation (Pg); Pt and Pg in 2019 were 522.1 and 442.3 mm, respectively, similar to the multiyear (1999–2018) means (507.9 mm yr-1 for Pt and 407.5 mm yr-1 for Pg). Seasonal variation was significant for 2019 precipitation, with 74.9 % occurring from June to September, according with the seasonal distribution characteristics of precipitation in the Loess Plateau. In this way, the year of 2019 was considered a normal precipitation year. Additionally, the highest monthly precipitation and highest single precipitation event occurred in August and September, being 128.3 and 57.3 mm, respectively.

Apple trees had dimorphic rooting systems, with fine roots distributed in shallow- and deep-layer soils (Fig. 3). The 11- and 17-year-old apple trees had similar root distribution profiles, with FRLD decreasing with increasing soil depth and the maximum FRLD occurring at 40 and 60 cm depth, respectively. Overall, 17-year-old apple trees had more fine roots in the whole profile than 11-year-old trees. Similarly, the SWC values of the two apple orchards had similar temporal and spatial variations in the profile (Fig. 4). The SWC was highly variable in the 0–100 cm soil layer during the sampling period due to rainfall infiltration, soil evaporation, and root water uptake but was relatively stable in the 240–500 cm layer. The 17-year-old orchard had greater seasonal variations and lower mean SWC values than the 11-year-old orchard (Fig. S2).

Root water uptake dynamics of apple trees are measured using the dynamics of δD values in xylem water following labeling (Fig. 5). During the BYF stage, no tracer signal was detected at 4 m depth, while pronounced uptake of artificial tracer occurred at depths of 1, 2, and 3 m for both apple orchards. During the FSW stage, tracer signals occurred in both apple orchards. Specifically, the 11- and 17-year-old apple trees quickly absorbed water from the 1 m soil layer, reflected in the markedly elevated δD value for the xylem samples taken on day 1 after D2O injections. The maximum concentration of D in xylem for both apple orchards occurred on day 3 after labeling, then decreased rapidly, and was lower than background values on day 7. Moreover, the presence of artificial D was found in the 17-year-old orchard at 2 and 3 m labeling depth, with the tracer peak occurring on day 3 after labeling, while none of the sampled trees in the 11-year-old orchard extracted water from soil profiles labeled at 2 m or deeper. During the FTM stage, both orchards had similar tracer uptake patterns to the FSW stage, with the peak δD value in xylem water occurring on day 3 or 5 after labeling.

The δD and δ18O values in soil water from both apple orchards had similar seasonal and vertical variations (Fig. 6). Shallower soils had more enriched isotopic values than deeper soils. The δD and δ18O values in the 0–40 cm soil layer varied significantly between sampling dates, attributed to the shallow infiltration of rainfall with negative isotopic values and intense surface evaporation. The isotopic values in the 100–500 cm soil layer did not significantly differ between sampling dates (P>0.05).

Isotopic values in xylem water depended on growing season stage and stand age (Fig. S3). Specifically, the BYF stage had more depleted δD and δ18O values for 11- and 17-year-old apple trees than the FSW or FTM stage. The similar isotopic values for xylem water in apple trees may be due to the same or similar water sources, with different isotopic values signifying different water sources.

The soil layer depths from which apple trees derived their water source during different growth stages were determined using isotopic labeling, before calculating the contribution proportion of different water sources to xylem water using the Bayesian mixing model. As shown in Fig. 7, isotopic values in xylem and soil water followed a similar trend to the local meteoric water line (LMWL). Across all samples, most of these isotopic values were plotted to the right of the LMWL, indicating that soil water in the study area came from precipitation and experienced intense evaporation. The relationship between isotopic values in xylem water and soil water revealed significant variation with growth stage and stand age, indicating that apple trees could extract water from different soil layers (Figs. 7 and 8). Specifically, the BYF stage produced more negative isotopic values in xylem water for 11- and 17-year-old apple trees (Fig. 7), which mainly used water from the 140–320 cm soil layer (more than 48 %) (Fig. 8). However, as precipitation infiltrated into subsurface layers, more positive isotopic values in xylem water occurred during the FSW and FTM stages. The isotopic values in xylem water were similar to it in the 0–40 cm soil layer (Fig. 7). Results from the mixing model revealed that the contribution of the 0–40 cm soil water reached 70 % for 11-year-old apple trees during the FSW stage and 49 % for 17-year-old apple trees during the FTM stage (Fig. 8). Overall, older apple trees relied more on deeper soil water during the growing season (Fig. 8).

The relationship between the contribution of water sources and SWC in apple orchards depended on soil depth (Fig. 9). When the data sets for a given soil layer were pooled, a significant (P<0.05) and positive relationship occurred between SWC and its contribution in the 0–40 cm soil layer, indicating that apple trees increased the utilization proportion of deeper soil water as SWC decreased in the 0–40 cm soil layer. It is worth noting that the relationship also depended on stand age. Specifically, a significant relationship in the 0–40 cm soil layer occurred in the 11-year-old apple orchard but not in the 17-year-old apple orchard. In addition, after pooling the data sets for a given soil layer, there was a significant (P<0.05) relationship between SWC and its contribution in the 140–320 cm soil layer.

Accurately determining the soil layer depths from which plants derive their water source is essential for quantifying their water use strategy in DVZ regions. However, it is often difficult to effectively capture changes in this depth caused by precipitation infiltration, affecting our evaluation of the contribution of different water sources to plant transpiration. In this study, we artificially injected D2O tracer to different soil layers to change soil water isotopic composition; then, root water uptake dynamics in the profile could be acquired by monitoring the concentrations of D in xylem samples, providing direct evidence for accurately determining root water uptake (Couvreur et al., 2020; Mennekes et al., 2021).

None of the apple trees in either orchard extracted water from the 4 m labeling depth during the BYF stage, but significant uptake of artificial tracer occurred at 1, 2, and 3 m depths (Fig. 5). This indicates that root water uptake in the 0–4 m soil layer is likely to meet the apple tree's transpiration, despite recent studies showing that the rooting depth of mature apple trees exceeded 5 m (Li et al., 2019; Wang et al., 2021c). We observed a distinctly different absorption of D2O tracer at different labeling depths in two orchards during the FSW stage (Fig. 5). At the 1 m labeling depth, an immediate reaction to D2O tracer was observed for the 11- and 17-year-old apple trees, in contrast to Evaristo et al. (2019) and Magh et al. (2020), who reported notable delays (days to weeks) between tracer injection and detection within xylem samples. This could be ascribed to the relatively tall canopy (>20 m) of their research trees, increasing the time taken for water to travel from roots to crown branches relative to apple trees. Unlike previous studies (Kahmen et al., 2021; Seeger and Weiler, 2021), the peak tracer concentration occurred on day 3 and then rapidly declined until day 7, possibly due to the heavy precipitation event with more depleted isotope on day 3. At the 2 and 3 m labeling depths, D2O tracer was found in the 17-year-old orchard, while none of the 11-year-old sampled trees extracted water from soil profiles labeled at 2 m and below (Fig. 5). This finding indicates that the 17-year-old apple trees absorbed more water from deep soils than the 11-year-old trees, consistent with the findings of Wang et al. (2020). The FTM stage had similar labeling results to the FSW stage, where xylem water in 11-year-old trees had more enriched isotopes at 1 m labeling depth and more depleted isotopes at 2 m and below than 17-year-old trees.

Numerous evidence from various ecosystems suggests that trees can adjust their water sources to adapt to changes in the surrounding water environment (Barbeta et al., 2015; Gao et al., 2018a; Ma et al., 2021; Zhao et al., 2021). In DVZ regions such as the Loess Plateau, soil water from precipitation is the primary water source for plant transpiration (Gao et al., 2021a; Tao et al., 2021a; Wu et al., 2022). Severe drought and water shortages and intense seasonal precipitation variation result in complicated and volatile soil water environments in this region. Hence, it is vital that apple trees adapt to this environment to survive and grow.

The contribution proportion of different soil water sources to plant transpiration was calculated using a Bayesian mixing model based on the depth of the tree's root water uptake determined using isotopic labeling. The isotopic signatures in xylem water and soil water depended on growth stage and stand age, such that water uptake patterns differed between apple tree growth stage and stand age (Fig. 7). The results from the Bayesian mixing model confirmed this finding (Fig. 8). More specifically, 11- and 17-year-old apple trees exhibited flexible water use strategies, shifting the main water source from deep- to shallow-layer soils as the growing season progressed (Fig. 8). This is consistent with several recent studies in this region (Huo et al., 2020; Zhao et al., 2020), in which root systems enabled trees to exhibit seasonal water use patterns, switching their water source between soil layers based on available soil water. Notably, SWC of the 0–40 cm soil layer significantly increased in apple orchards due to precipitation recharge during the FSW stage (P<0.01) (Fig. 4), and 11-year-old trees rapidly changed their main water source to shallow soil layer (0–40 cm) (70 %) in this stage, while 17-year-old trees changed their main water source to shallow soil layers (49 %) until the FTM stage (Fig. 8). One possible explanation is that drought and high temperatures before sampling caused a less reversible embolism, reducing the water conductivity of the shallow root system (Grossiord et al., 2017). This difference indicates that the root water uptake response to soil water change depends on stand age, with 11-year-old trees more sensitive than 17-year-old trees in the 0–40 cm soil layer, and it is verified by the relationship between the contribution of water sources and SWC in the 0–40 cm soil layer (Fig. 9). Similarly, Huo et al. (2018) and Wang et al. (2021a) reported that old trees tended to access stable deep water sources and had a time lag converting water sources following a soil water change. Overall, the apple trees in both orchards absorbed the highest proportion of water in the 140–320 cm soil layer during the BYF stage. Moreover, 17-year-old apple trees had a higher proportion of water from these layers than 11-year-old trees throughout the growing season.

Plant root water uptake is sensitive to changes in the water environment, and changes in water uptake strategy affect their ecosystem functioning. In DVZ regions, accurately calculating the contribution proportion of different water sources to transpiration helps us to understand the variation in plant water use strategies. However, few studies have determined the soil layer depths from which trees derive their water source when calculating the contribution of water sources based on the natural abundance of stable isotopes method. Numerous studies have determined this depth based on prior information (e.g., vertical root distribution) (Tao et al., 2021b; Wu et al., 2022; Zhao et al., 2021), which can reach 5 m to even 10 m. The existence of roots in deep soils does not necessarily mean that root water uptake occurs (Szutu and Papuga, 2019). The isotopic labeling results showed that no water uptake occurred in deep (>140 cm) soils for 11-year-old apple trees during the FSW stage (Fig. 5). Thus, quantifying the water use strategy of trees in DVZ regions using the natural abundance of stable water isotope method may not be accurate since this method assumes that trees always extract shallow- and deep-layer soil water. This study used isotopic labeling in deep layers to identify whether trees absorb DLSW and determine the soil layer depths from which trees derive their water source. Then we calculated water sources based on the natural abundance of stable isotopes method, and compared the results with the natural abundance of stable water isotopes method (Figs. 8 and S4). The results showed that the natural abundance of stable water isotopes method overestimated the contribution proportion of DLSW (>140 cm) for 11-year-old apple trees (17 %) during the FSW stage. The root water uptake depth from isotopic labeling indicated that soil water in the 0–140 cm layer replenished by precipitation could meet the transpiration demand of apple trees during the FSW stage (Figs. 5 and 8), improving our understanding of the relationship between the contribution of water sources and SWC. The contribution proportion of water sources in the 0–40 cm soil layer increased significantly with SWC in the 11-year-old apple orchard (P<0.001), similar to the findings of previous studies (Gao et al., 2018b; Grossiord et al., 2017). In contrast, no significant differences occurred between the contribution proportions from the 2.4–3.2 and 2.4–5 m soil layers in the 17-year-old apple orchard (Figs. 8 and S4), and trees did not actually acquire water from the 3.2–5 m soil layer. This further indicates that determining the soil layer depths from which trees derive their water source is important for understanding the role of DLSW on plants, especially those with flexible water uptake strategies.

The apple trees on the Loess Plateau are heavily dependent on soil water in deep layers due to low annual precipitation (400–600 mm) and abundant soil water resource in the DVZ (Wang et al., 2021c; Yang et al., 2022). Our results show that apple trees switch their water sources between different soil layers to adapt to the changing water environments on the Loess Plateau, which is particularly important in the context of climate change. Special attention should be directed to water consumption in deep soils – we found that apple trees absorbed the most water from deep soils during the BYF stage, with 17-year-old apple trees consuming more water in these layers than 11-year-old trees throughout the growing season. This result is in accordance with previous observations in this region that soil water availability gradually decreased with increasing stand age, and then apple trees absorbed more water from deeper soil layers (Li et al., 2019). Similarly, Barbeta et al. (2015) found that trees increased their use proportion of deep soil water and groundwater following a long-term (12 years) experimental drought. However, this water use strategy may not be sustainable for trees in DVZ regions where deep soil water is difficult to be recharged. The result of the tritium peak method suggested that it took more than 50 years for soil water migration to 6 m depth in apple orchards in DVZ regions (Li et al., 2018). Thus, once DLSW is depleted, it cannot be replenished within a short timeframe, reducing the tree's ability to resist water stress. Also, Wu et al. (2021) observed that soil water generated by precipitation was the primary water source for apple trees when DLSW was depleted, dominating their transpiration. In this case, trees were likely to encounter irreversible embolism, increasing the risk of drought-induced mortality, threatening the sustainable development of vegetation and changing the regional hydrological cycle (Brodribb et al., 2020; Zhang et al., 2020). Therefore, we suggest long-term and high-frequency monitoring of isotopes in soil and xylem water, especially at a large geographical scale, to further understand the long-term changes in plant water use strategy and evaluate their adaptability under climate change. In addition, we observed the rapid appearance of D signals in xylem following tracer injection, indicating the exchange of bound and mobile water pools in soil and challenging the “two-water-world” hypothesis. These findings provide an important reference for evaluating the validity of this hypothesis in semi-arid areas.

In this study, isotopic offset between xylem and soil water was observed for both 11- and 17-year-old unlabeled apple trees (Fig. 7 and Table S2). We used the isotopic composition of soil water to correct δD values of xylem water, ensuring they match those of soil water. Although we did not collect soil water isotope samples in the isotopic labeling experiments, this may have little effect on determining the soil layer depths from which trees derive their water source due to the high δD values in the injected solution. It should be noted that isotopic spatial heterogeneity of xylem water induced by sampling position and time (Nehemy et al., 2022) and soil water induced by uneven distribution of throughfall and preferential flow (Xiang et al., 2019; Yang and Fu, 2017) could lead to an isotopic offset. Recently, isotopic offsets between plants and their potential water sources have also been found in various ecosystems, which may hinder the unambiguous identification of water sources and influence the accurate assessment of DLSW utilization (Barbeta et al., 2022; de La Casa et al., 2022; Zhao et al., 2016). Some studies argued that isotopic fractionation during root water uptake could be attributed to the existence of Casparian strips which can lead to isotope enrichment in root water and depletion in xylem water (Naseer et al., 2012; Vargas et al., 2017). Seeger and Weiler (2021) questioned whether xylem water was completely renewed by newly absorbed soil water, thus affecting isotopic offset. Furthermore, CVD may mask or exaggerate the isotopic offset, although it was the most common methodology (Chen et al., 2020; Orlowski et al., 2016a, b, 2018; Wen et al., 2022). When quantifying the water use strategies of plants, the isotopic measurement bias related to CVD should be considered. As a whole, there are various trends and causes of isotopic offset. Further research about offset is urgently needed to better understand root water uptake processes.