The study site was located in Yan’an City, Shaanxi Province, China, in the Yangjuangou catchment (36°42′N, 109°31′E, 1298 m in elevation) ( Figure 1 ), where 531 mm of long-term mean annual rain occurred (from 1952-2012) and the mean temperature was 9.4 °C (Liu et al., 2023a). The original woody species were scarcely present because of intensive human activities. The site is dominated by the typical revegetated shrub species Vitex negundo, which is a perennial deciduous shrub that has been widely planted for ecological restoration since 1999 (Yuan et al., 2022), and the study plot experienced anthropogenic disturbances ( Figure 1D ). The height of V. negundo ranged from 1.4 to 2.2 m, with an average of 1.83 m. The sample plot was mainly composed of 31 scattered stems, and the stem diameter ranged from ranged from 12.54 to 25.61 mm, with average of 18.38 mm. The stems of diameter 15-24 mm, representing 80.65% of the total stems ( Figure 1E ), must have been responsible for the major part of the water use of V. negundo. The stems equipped with sap flow gauges were selected in this intermediated size range to represent of the most significant proportion of the stem population ( Table 1 ).

The En data of V. negundo individual stems were measured using the heat balance method because of its accuracy for shrub stems (Dugas et al., 1993; Hall et al., 1998). Different types of sensors were used according to the basal diameter of the stems (Dynamax Inc., Houston, TX, USA; Model SGA 13, and SGB 19). The theory and methodology of this method have been described in detail in previously study (Yue et al., 2008). Measurements were replicated using three individuals of V. negundo. The sap flow sensors were installed in each stem at the base (0.2 m above the ground) and at the top (the section of the canopy branch) of the live crown ( Figure 1B ). Data were sampled every 10 s, averaged and recorded at 30 min intervals. Measurements of sap flow density were made from June to September 2021 and 2022. Measurements began 2 days after the sensors were installed for data stability in June and continued until the end of September. Because the sap flux is correlated with tree size (Wang et al., 2012), daily sap flux of the sample stems was normalized with the stem area (g h^-1 cm^-2).

The stem water potential (Ψs, MPa) was automatically measured with thermocouple stem psychrometer (PSY, ICT International Pty., Australia). Thermocouple stem psychrometer could continuously record changes in plant water status, directly reflecting the energy needed by the plant to obtain water or the stress on the plant. According to the distribution of stem diameter in the sample plot, an ideal sample stem were selected with 18.42 mm diameter, representing the mean stem diameter of sample plot. The psychrometer was installed at 160 cm above the ground to measure the xylem water potential ( Figure 1C ). Before the instrument was installed, an installation section of 1cm wide and 5-6 cm long of stem was scraped with a knife, which need connect to the psychrometer. It was rinsed 3-5 times using deionized water and then rubbed with paper towels to ensure complete drying. A silicone grease seal was used for the connection between the psychrometer and stem. The gauge was protected using a shield wrapping of several layers of aluminum foil to reduce the effect of weather. The data were recorded at 15-min intervals in 2022.

The quantities of daily and hourly En and Qn were the products of the nighttime sap flux density at the base and top at daily and hourly times, respectively. If the quantity of stored water in the crown was lower than that in the stem, the sap flux at the top was taken as Qn (Daley and Phillips, 2006), while the sap flux at the base represented En. The values of stem Rn were defined as the discrepancies between En and Qn. Negative Rn values indicated water withdrawal, while positive Rn values indicated excess water refilling depleted water stores. The sums of instantaneous changes in stored water were calculated for daytime and nighttime on hourly and daily time scales during the 2021 and 2022 from June to September.

An automatic weather station was installed in the open area of the catchment at approximately 100 meters away from the experimental field and 1.7 meters above the ground ( Figure 1A ). Air temperature (Ta, °C) and relatively humidity (RH, %) were measured using a HMP155 sensor (Vaisala, Finland). Solar radiation (Rs, W m^-2) was monitored using a pyranometer (Model CNR 4, Kipp& Zonen B. V, the Netherlands). All climatic data were measured once per minute and recorded every 30 minutes from June 1 to September 30. Rainfall amount (mm) was recorded using a tipping-bucket rain gauge (TBRG) (Model RG3-M, Onset Computer Corporation, USA) mounted. The vapor pressure deficit (VPD, kPa) was calculated based on a Ta and RH equation (Campbell and Norman, 1998). Volumetric soil water content (SWC, %) at the study site was measured using EC-5 soil moisture probes (Decagon Devices Inc., Pullman, WA, USA) that were installed at seven depths below the soil surface (10, 20, 40, 80, 120, 160, and 200 cm). The environmental data were sampled and recorded at the same frequency as sap flow measurements.

To provide a conservative estimate of when stem refilling ceased, we used the time when radiation became <5 W m^-2 or zero to define the beginning of the ‘night-time’ period in this study.

To identify the relationships between En components and environmental factors, between Rn and daytime sap flow and predawn Ψs, general linear regressions (y=a+b·x) were fitted to the data. Regression slopes were used as an indicator of overall sensitivity of Qn to the variation in biophysical variables. We applied one-way ANOVA to test the statistical significance of differences between Qn and Rn at a critical probability value of 0.05. Pearson’s correlation coefficient was used to estimate associated pairwise relationships between all variables. Stepwise multiple regression analysis was used to examine the contribution of Ta, RH and SWC in different soil layers to explaining the variations of Qn. All statistical analyses were performed with the SPSS software package (Version 19.0 for windows, SPSS Inc., USA). Linear fits were performed in Origin (Version 8.0, OriginLab Corp., USA).

In order to minimize the effect of rain events on nighttime sap flow measurements, only data collected during the non-rainy parts of the growing season were included in the analysis (Hayat et al., 2021).

The nighttime VPD was higher in 2021 than in 2022 ( Figure 2A ), while the opposite trend was observed for RH (p<0.01) ( Figure 2C ). Interannual variations in Ta and Rs exhibited no significant differences between the two years (p>0.05), with interannual means (± standard deviation) of 21.79 (± 0.29)°C, 202.68 (± 6.10) W m⁻², respectively ( Figure 2B, D ). These hydrometeorological variables were lower at night than during the day (p<0.05). The total precipitation during the study period (June to September) was 371.40 mm in 2021 (a normal year), 450.87 mm in 2022 (a wetter year, 19.88% wetter than long-term average value (376.11 mm from 1952-2012)). Ta at night showed significant differences between the two years (p=0.03) ( Figure 2B ). In general, the SWC in the three soil layers (0-40 cm, 40-120 cm, 120-200 cm) in 2022 were significantly greater than those in 2021 ( Figure 2E ) (p<0.05).

The values of Qn and Rn fluctuated during the growing season ( Figure 3 ). The daily whole-stem Qn values during clear days ranged from 0 to 41.45 and 0 to 52.94 g cm⁻² over the study period, with means of 4.68 (± 0.59) and 10.53 (± 1.31) g cm^-2 per day in the main growing seasons in 2021 and 2022, respectively. The mean values of Rn were 0.86 (± 0.14) and 2.02 (± 0.27) g cm^-2, ranging from 0 to 5.84 and 0 to 9.80 g cm^-2 in 2021 and 2022, respectively.

On monthly and interannual time scales, Qn was significantly greater than Rn (p<0.05). Qn and Rn were both significantly greater in 2022 than in 2021 (p<0.01), except for Qn in July. The maximum values of Qn and Rn were 15.89 g cm^-2 day^-1 in Sep 2022 and 2.55 g cm^-2 day^-1 in June 2022, respectively. The monthly mean values of Qn and Rn were 7.92 and 1.29 g day^-1 cm^-2 in 2021, 9.62 and 2.18 g day^-1 cm^-2 in 2022. The ratio of Qn/En was lower in 2021 (79.76%) than in 2022 (83.91%) (p=0.03). The average values of Qn/En and Rn/En in in two years were 81.83% and 18.17%, respectively ( Figure 4 ).

The time series for five days from July 4 to 8 in 2022 were chosen to provide typical basal and crown En and Ψs patterns on clear sunny days. During the morning hours, the water balance of the shrub stem between the base and top was positive, the inflow from the lowermost stem was lower than the outflow to the upper stem ( Figure 5 ), and Ψs was relatively high at predawn. The balance became negative when the input into the stem was lower than the output into the stem until approximately after 9 p.m. with decreasing Ψs. Plant depleted water storage was recharged at night until the early morning hours of the next day transpiration resumed. En returned to almost zero before sunrise. The Ψs decreased from sunrise until the afternoon at approximately 6 p.m., with a minimum of -1.24 MPa ( Figure 5B ). Then, Ψs increased until predawn, corresponding to the recharge period of the stem.

In our experiment, the mean hourly Qn and Rn values were 1.65 and 0.31 g cm^-2, respectively, and the two fluxes were both greater in the wet year. The mean percentage of Qn to En in this study was 81.83%. Consequently, Qn was the predominant contributor to En during the study period. Although the percentages of En to daytime sap flow were greater in water-limited regions (Snyder et al., 2003; Wang et al., 2012), the proportions of Qn and Rn to En did not reach the same conclusion in previous studies. Some studies have shown that En is mainly used for refilling water in trunks. Researchers have shown that the percentage of Rn to En was 40%-70% for Quercus douglasii (Fisher et al., 2007), ranging from 14.7 to 30.3% for Acacia mangium (Wang et al., 2012), and approximately 90% for Acer truncatum (Wu et al., 2020), indicating a greater allocation of En to Rn. In temperate woodlands, Qn/En was 50-70% for two co-occurring evergreen species (Eucalyptus parramattensis and Angophora bakeri) (Zeppel et al., 2010). Studies showed that Rn was strongly affected by tree features, with the exception of environmental factors, such as plant size (Horna et al., 2011); plant sapwood (Carrasco et al., 2015); basal area (Wang et al., 2012) and plant canopy (Williams et al., 2021), which exhibited greater capacitance. Because of the significance of Rn for daytime water loss, the amount of water stored is greater for trees than for shrubs. The volume of leaves and small branches above the top sensor were not considered, which may lead to underestimation of stem refilling volume in this study.

Studies have observed that stem refilling for Eucalyptus saligna finishes by 23:00 h because daytime water use causes a deficit in the internal water storage of plants (Daley and Phillips, 2006; Zeppel et al., 2011), which can be recharged through nighttime stem refilling for tree species. However, in the present study, the refilling of capacitors increased from dusk to predawn, while Ψs stopped increasing until predawn, suggesting that the nocturnal recharge of stem water storage was essentially complete during the night ( Figure 5 ). The correlation between Qn and Rn and its variations suggested that water recharge and transpiration were synchronous at night ( Table 2 , Figure 7 ). Research has also shown that there is no discernible distinction between Qn and Rn during the process of En (Daley and Phillips, 2006; Huang et al., 2017). Therefore, using only the time to distinguish Qn from Rn would introduce error.

Variation in the components of En may be an adaptation in response to resource deficiencies and may provide eco-physiological advantages for plant growth. In this study, En is mainly used for Qn. Thus, in species like V. negundo, increased evaporation demand during the night may affect the plant water use and ecological water balance of plant ecosystem (Lombardozzi et al., 2017). Studies also proposed the Qn may improve oxygen supply to the sap wood or prevent of CO₂ build-up in leaves during the night (Marks and Lechowicz, 2007), reduce leaf surface temperature (Peraudeau et al., 2015). It may also transport nutrients to the roots and distal parts of the plant, which is important in nutrient-limited but water-sufficient areas (Scholz et al., 2007). In water-limited areas, Qn may reduce the water potential of plant leaves and inhibit hydraulic redistribution.

Although Rn was a small fraction of En in this study, the predawn Ψs in 2022 and the daytime sap flow in both years were positively correlated with Rn (R² = 0.38, p<0.01 for predawn Ψs; R² = 0.31, p<0.01 for daytime sap flow) ( Figure 8 ), which may facilitate stomatal opening and subsequent carbon fixation during the early morning. Our results also showed that En was correlated with daytime sap flux ( Figure 8B ) (Wang et al., 2012), indicating that an amount of nighttime sap flux was used for stem water recharge as a result of high water loss during the day (Huang et al., 2017). The mean Rn contributed 18.17% of the En in this study suggesting the importance of Rn for the adaptive strategy of shrubs to soil water scarcity. The contribution of water storage in the stem to transpiration accounted for 10-20% of the daily transpiration for Japanese red pine and an oak forest (Kobayashi and Tanaka, 2001). Therefore, an estimation of Rn could provide an in-depth understanding of plant adaptation to drought stress. If the VPD_night increased with the warmer and drier condition under climate change, the percentage of Rn will continue to increase with Qn, which would change the adaptability to the environment. Therefore, the occurrence and amount of Rn are more important for plant survival, especially in water-limited regions.

There are some limitations in this study, including the low sample replication and systematic errors of gauge heater. The replication rate of this study for the sample size may not represent all the morphological and physiological features of the species and there is a possibility of overlooking some specific stem sap flow information. Due to field conditions and equipment limitations, 3-6 samples have been commonly used in previous studies for monitoring plant sap flow or transpiration, such as three samples (Yu et al., 2016, 2018), four samples (Wu et al., 2020) and six samples (Yue et al., 2008). On the other hand, the presence of measurement instrument errors would be another limitation in this study. Although the results may be acceptable using this method (Flo et al., 2019), the systematic errors in the heat balance method were not considered in this study, such as radial variations in sap flow, species specific differences in the parameters setting in the instrument (Moro et al., 2004; Repo et al., 2008) and the environmental conditions that cause temperature differences along the stem (Shackel et al., 1992). These uncertainties need to be improved in future studies.