Evapotranspiration of xerophytic shrub Salsola passerina and Reaumuria soongorica in an arid desert ecosystem of NW China

Evapotranspiration (ET), including soil evaporation (E) and plant transpiration (T), is a fundamental process of ecological, hydrological and atmospheric systems and a major component of water balance especially in the desert area, where ET accounts for more than 95% of precipitation. The controlling factors of ET vary in different ecosystems, and are dynamic across intra-annual to decade time scales (Wilcox et al. 2003; Huxman et al. 2005; Ryu et al. 2008). ET is one of the most problematic components of the hydrological cycle, and is one of the most difficult variables to accurately quantify since it exhibits high spatial and temporal variability (Gebler et al. 2015).

Numerous methods have been applied to measure or estimate ET. At regional or global scales, the remote sensing approach has been increasingly used to map the spatial distribution of ET (Kramber et al. 2010; Mariotto et al. 2011; Anderson et al. 2012a, 2012b; Wang et al. 2013; Xiong et al. 2015). Nevertheless, using the remote sensing method to accurately estimate ET in arid regions with large barren or sparsely vegetated areas is still a challenge for the areas characterized by high spatial heterogeneity of aerodynamic and land surface properties (Vinukollu et al. 2011; Tian et al. 2013). At scales smaller than 1 km², soil water balance, Bowen ratio energy balance (BREB), eddy covariance and the large weighing lysimeter technique had been widely used for measuring ET directly or indirectly (Allen et al. 2011a, 2011b). The advantages of the soil water balance method are that it is not limited by time and space scales and complies with the principle of water balance. Disadvantages and challenges are that deep percolation losses or gains are difficult to measure, and that differential spatial wetting of soil due to local spatial variation in irrigation (or precipitation) additions is difficult to be obtained (Cholpankulov et al. 2008). The BREB method is a practical and relatively reliable micrometeorological method with the advantage of low cost, and simple and straight forward measurements. The numerical instability of the BREB equation, however, is evident during periods of Bowen ratio values in the vicinity of −1, and heavy reliance of the ET estimate on the accuracy and representativeness of the net radiation and soil heat flux density measurements (Allen et al. 2011a). The eddy covariance method is increasingly used in ET measurement because of high frequency, non-destructive, continuous direct recording (30 min), automated systems, and the ability to measure radiation, latent, sensible heat flux and CO₂ flux simultaneously. Eddy covariance has disadvantages including complex instrumentation, energy balance closure error (approximately 10–30%), and underestimation of ET (Foken 2008; Ding et al. 2010).

A number of previous studies suggested that a large weighing lysimeter is the most accurate and reliable field method and the data are regarded as standard for actual ET measurements in desert areas which are used to provide baseline information for development, calibration, and validation of other measurement techniques (Payero & Irmak 2007; Goss & Ehlers 2009; Marin et al. 2010; Evett et al. 2012; Yang et al. 2014; Marek et al. 2016). Despite the rather high costs of installation and maintenance, boundary effects, limited areal extent (areas generally ranging from 0.05 to 40 m²), considerable effort for data processing, and extreme sensitivity to environmental factors, numerous lysimeter facilities exist worldwide (Wang et al. 2004b; Lanthaler 2005; Ruiz-Peñalver et al. 2015). Furthermore, a weighable lysimeter can not only register ET on a natural surface at ground level, but can also allow a precision determination of the mass change and outflow in high temporal resolution, and calculation of precipitation by rain, dew, fog, rime and snow in a hitherto unprecedented accuracy (Meissner et al. 2007; Vera-Repullo et al. 2015; Hoffmann et al. 2016). It should be noted that although lysimeter measurements are prone to random and systematic errors due to size and external factors like wind, heavy precipitation and animals etc. (Nolz et al. 2013; Schrader et al. 2013), random errors can be greatly reduced by using relatively large size lysimeters (e.g. ≥1 m²) (von Unold & Fank 2008), and systematic errors can be handled by filtering noisy lysimeter data (Vaughan & Ayars 2009; Peters et al. 2014, 2016; Hannes et al. 2015). Actually, lysimeters have been successfully used to obtain accurate ET rates for both short and long periods because of their high accuracy and time resolution (Allen et al. 1998), and to estimate water and solute fluxes directly at arid sites (Gee & Hillel 1988). Gee et al. (2009) have suggested that lysimeter and direct water balance measurement techniques are more reliable at arid sites than other estimation methods.

The desert ecosystems are sensitive to environmental changes, and have the potential for catastrophic changes (Newman et al. 2006). Desert ecosystems in China are characterized by vegetation patchiness, of which the vegetation patterns, composition and ecosystem processes are determined by hydrological processes (Li et al. 2013). Highly variable precipitation with discrete pulses characteristics is often the sole source of water in these desert ecosystems (Noy-Meir 1973; Loik et al. 2004; Magliano et al. 2015). Therefore, a quantitative understanding of the hydrologic balance induced by desert vegetation is vitally important for assessing the spatial patterns of vegetation and improving water use efficiency in the desert (Rietkerk Max et al. 2004; Wu et al. 2014). Furthermore, ET is of prime interest in the hydrologic balance for water allocation and vegetation water use (Kool et al. 2014). Quantifying ET of the dominant xerophytic shrub community in a desert ecosystem where the shrub mainly utilizes precipitation provides insights into hydrologic processes in desert regions. At present, however, the long-term observation of the ET of typical xerophytic shrubs like Reaumuria soongorica (C₃ plant) and Salsola passerina (C₄ plant) community, which are widely distributed across desert regions of China, has not been reported. R. soongorica and S. passerina may exist either in single shrub communities or in associated shrub communities and are considered as climax communities. One recent study indicated that the associated shrubs community of the two shrubs decrease the transpiration rate of the entire community (Su et al. 2012). This raises a question: of the single shrub community or the associated shrubs community, which plays a more important role in contributing to regional ET? We hypothesized that ET differs between the single shrub and associated shrubs community. If this is the case, how do morphological traits, rainfall, soil moisture characteristics and other meteorological variables affect ET variability? Therefore, we initiated the following specific objectives: (1) characterizing actual E and ET variations at daily and annual time scale by large automatic weighing lysimeters; (2) quantifying the influences of morphological traits, rainfall, soil moisture characteristics and other meteorological variables on E and ET. The R. soongorica and S. passerina xerophytic shrubs are regarded as climax communities of the study area; the present study on ET of those communities is expected to be used to assess the ecological water requirement and evaluate ET estimation models in arid desert ecosystems.

The experiments were conducted at the Shapotou Desert Experimental Research Station (SDERS) of the Chinese Academy of Sciences, which is located at the southeastern border of the Tengger Desert, NW China (37°32′N, 105°02′E) at an elevation of 1,339 m. The area was classified as ecotone between desertified steppe and sandy desert with an annual mean precipitation of 186 mm, approximately 80% of the rainfall events occurring between May and October in the growing season with a large inter-annual variability as high as 45.7%. Annual mean potential evaporation reached 3,000 mm (observed by an evaporation pan of type E-601). The groundwater table below the land surface is more than 60 m. Therefore, the effective water source for plants is primarily precipitation (Berndtsson & Chen 1994). The climate at the experimental site is characterized by an abundance of sunshine and low relative humidity. The minimum average monthly relative humidity is 32.7% during April, and the maximum is 54.9% during August. The sunshine hours range from 6.7 to 6.9 h per day from October to January, and from 7.0 to 8.2 h per day in the remaining months. The annual average temperature is 10.0 °, with historical extreme maximum and minimum temperatures of 38.1 ° in July and −25.1 ° in January, respectively, during the last 50 years (Gao et al. 2016). The average annual wind velocity is 3.5 m s⁻¹ (2 m height) and 122 d involve dust events (Liu et al. 2006).

Reaumuria soongorica (Pall.) Maxim and Salsola passerina Bunge are perennial extremely xerophytic shrubs and are considered as climax communities in the southeast edge of the Tengger Desert. The shrubs are constructive and dominant species of steppe and typical deserts and play a vital role in sustaining the ecological stability of the deserts due to their drought-resistant nature. R. soongorica is many-branched and ranges in height between 10 and 70 cm; characterized by short and cylindrical leaves with a length of 1–5 mm and width of 0.5–1 mm, and normally 4–6 clustered on shortened branches. In addition, there is a unique adaptive strategy for R. soongorica leaves to enter a state of dormancy during dehydration but revive when rewetted; roots are at depths between 10 and 100 cm and 93% of the feeder root area is typically maintained in the 0–60 cm depth (Xu & Li 2009; He et al. 2015). S. passerina is also many-branched with a height between 15 and 50 cm, with a 2–3 mm length and 2 mm width of cone or triangle shaped leaves and root depths between 0 and 50 cm (Yang et al. 2013).

A set of large automatic weighing lysimeters with accuracy of ±0.1000 kg (equivalent to 0.025 mm water of the 4 m² lysimeter) is established in the SDERS of the Chinese Academy of Sciences. The outer cylinder of the lysimeter was 250 cm in depth and 256 cm in diameter. The inner cylinder was 250 cm in depth and 226 cm in diameter (Figure 1).

The gap between the outer and inner cylinders was covered with polyurethane foam to prevent water, soil, and rodents from getting inside the outer box, as well as to help minimize heat transfer between the atmosphere and the airspace between the two cylinders. The lysimeter bottom was first covered with a 20 cm drainage layer graded from 1 to 2 cm gravel, serving as a capillary break. Then the local fine dune sands were repacked with a depth of 220 cm, which had an initial average volumetric soil water content of 2.5%, and the volumetric field capacity of the dune sand was about 6.7% (Wang et al. 2004a). Each lysimeter had two heat flux plates 80 mm in diameter (HFP01SC, Hukseflux Thermal Sensors, Delft, The Netherlands) at 5 and 7 cm below soil surface, respectively. The Stevens pF Sensors (for measuring soil matric potential and soil temperature) and TDR probes (for measuring water content) were installed at soil depths of 10, 20, 40, 60, 80, 100, 160, 180, 200, and 220 cm within each lysimeter. In addition, two samplers (flexible polymer tube) were installed at soil depths of 100 and 150 cm for percolation collection which was extracted by a vacuum pump.

Four electronic auto weighing lysimeters were used to determine actual evaporation from a bare ground surface and actual ET of xerophytic shrub from a vegetated area under natural precipitation. Each vegetated lysimeter transplanted four xerophytic shrubs (Figure 1).

The root depth was determined during the transplantation process, and it was found that both the experimental shrubs are shallow-rooted with a root depth varying between 0 and 60 cm, which concentrates within the soil profile at around 40 cm depth. The morphological traits of vegetated shrubs in lysimeters are shown in Table 1. The shrub height and new shoot length were measured with a ruler, and the Canopy projection area (approximated as an ellipse) was determined by measuring the longer and shorter sides of the plant canopy, and then calculated by the formula of the ellipse area. The relative cover of vegetation was estimated by canopy projection area and lysimeter area. The leaf area index (LAI) was measured directly by a LAI-2000 plant canopy analyzer (LI-COR, Inc., Lincoln, NE, USA). The measurements were carried out at the end of each month during the growing season in 2015 and 2016.

Changes in lysimeter weight are used to quantify loss (E and ET) of water from the lysimeter soil monolith, measured by a set of three electronic sensors (load cell) for each lysimeter. The lysimeter mass change in kg for every 30 minutes is converted to a mass equivalent relative lysimeter storage value (mm of water) by dividing it by the cross-section area of the lysimeter (4 m²) and the density of water (taken as 1,000 kg m⁻³) (Marek et al. 2016).

Details of measuring event-based precipitation and meteorological variables were demonstrated by Wang et al. (2016) and Zhang et al. (2015). In order to understand the change of actual E and ET in accordance with rain depth and distribution of precipitation based on the daily time scale in the given years of 2015 and 2016, we assigned an individual day of recorded precipitation to a rain event.

The year 2016 can be considered a relatively wet year compared to the mean annual precipitation of 186 mm, since 42 mm precipitation was artificially added to the four lysimeters, with 15, 6 and 21 mm added to the four lysimeters on 20 May, 27 May and 29 June in 2016, respectively. The added precipitation was only excluded in Table 2 to analyze natural precipitation distribution.

The annual precipitation was 140 and 171 mm in 2015 and 2016, with 85 and 89%, respectively, i.e. 119 and 151 mm falling during the growing season from May to October of the experimental years 2015 and 2016, respectively. Apparently, the monthly precipitation distribution was fairly heterogeneous during the growing season of 2015 and 2016. A total of 53 and 45 events occurred in 2015 and 2016, respectively. The precipitation events with 0–5 mm accounted for 89 and 80% of annual total events in 2015 and 2016, followed by 5–10 and >10 mm rainfall events, which accounted for 8 and 4% in 2015, respectively, and 13 and 7% in 2016, respectively (Table 2).

Figure 2 shows the seasonal patterns of E and ET in the growing season of 2015 and 2016. Arrival times of precipitation events were consistent with periods of relatively high E and ET of the bare area and xerophytic shrub community, especially under large precipitation events. The daily E of bare lysimeter increased from 0.1 to 2.3 mm/d, the daily ET of S. passerine (ET_S) increased from 0.3 to 2.3 mm/d, the daily ET of R. soongorica (ET_R) increased from 0.3 to 2.5 mm/d, the daily ET of mixed community of S. passerine and R. soongorica (ET_S+R) increased from 0.3 to 2.5 mm/d at the end of the large precipitation event with a total amount of 24.8 mm on 20 May, 2015. In the growing season of 2015, ET_S and ET_R had the maximum values of 138 and 138 mm, respectively, followed by ET_S+R (128 mm) and E (114 mm). In contrast, in the growing season of 2016, ET_S had the maximum value (221 mm), followed by ET_R (198 mm), ET_S+R (189 mm), and E (170 mm). Total E, ET_S, ET_R, and ET_S+R of the growing season increased by 49, 61, 44 and 47%, respectively, from 2015 to 2016. It can be seen that E was lower than the corresponding precipitation, with a low soil evaporation to precipitation ratio (E/P) of 0.95 and 0.88 in 2015 and 2016, respectively. ET of xerophytic shrubs exceeded the corresponding precipitation, except the mixed community of S. passerine and R. soongorica during the growing season of 2016, where the ET to precipitation ratio (ET_S+R/P) was 0.98. During the growing season of 2015, the ET_S/P, ET_R/P and ET_S+R /P ratio were 1.15, 1.15, and 1.07 in 2015, respectively. The ET_S/P and ET_R/P were 1.14 and 1.02 in 2016, respectively. The daily average E and ET_S, ET_R, ET_S+R were 0.62, 0.75, 0.75 and 0.70 mm d⁻¹, respectively, in 2015; they were 0.93, 1.20, 1.08 and 1.03 mm d⁻¹, respectively, in 2016. It is clear that daily average E, ET_S, ET_R, and ET_S+R increased by 50, 60, 44 and 47%, respectively, from 2015 to 2016.

The daily E, ET_S, ET_R, and ET_S+R after a large rainfall event (20 May 2015) were selected to characterize their response to precipitation during certain time intervals (8 days). The daily E, ET_S, ET_R, and ET_S+R increased in accordance with rainfall (20 May) and showed a sharp decline on the second day (21 May) and then slowly declined (Figure 3). They had the minimum values on the sixth day (25 May), and after that there was a slight increase.

There was a significant linear relationship between the accumulative E or ET and the accumulative P in the growing season of 2015 (Figure 4(a)) and 2016 (Figure 4(b)), respectively. The percentages of precipitation explanation for E and ET were more than 98%. Hence, E and ET were dominated by rainfall patterns in the period from May to October. The slope of the linear regression of ET on P is larger than that of E on P.

As seen from Figure 5, E and ET had a significant positive correlation with wind speed, solar radiation and air temperature, but a significant negative correlation with relative humidity on the daily time scale during the growing season. Wind speed is the main factor affecting E and ET on a short time scale during the growing season, interpreting around 72% of E and 71% of ET variation, which was higher than other meteorological factors. Although the daily E, ET_S, ET_R, and ET_S+R after a large rainfall event within 8 days were significantly correlated to all meteorological factors, it was more affected by the relative humidity compared with the other climatic variables (Table 3).

Differences in soil water content from the consecutive soil layers within four experimental lysimeters were used to determine the wetting front. One relatively large rainfall event with a cumulative rainfall of 24.8 mm started from 8:15 am (0 h) and ended at 22:00 pm (14 h) on 20th May in 2015 and was selected for the analysis (Figure 6). The wetting front reached a maximum depth of 40 cm, with no noticeable changes found in soil moisture at 60 and 80 cm depth within 8 days (Figure 6(a)–6(h)). The soil moisture at the upper soil profiles of 10 and 20 cm in all lysimeters increased to a maximum on the following day (Figure 6(b)) after the rainfall event and then declined slowly. In contrast, the change in soil moisture at a depth of 40 cm lagged more than 24 h behind that of the upper soil layers (Figure 6(c)). Figure 6 also shows that the wetting front advances are similar in all lysimeters after rainfall, but differences are shown at profiles of 20 cm. For example, soil moisture content was higher at the 20 cm depth of vegetated lysimeters than that of non-vegetation lysimeters (Figure 6(b)–6(h)). Soil moisture content was also higher at 40 cm depth of lysimeters of single S. passerina (Figure 6(c) and 6(d)) and the associated shrubs community (Figure 6(c)–6(d)).

As seen from Table 4, E was significantly correlated with soil moisture content at depths of 10 and 20 cm (P < 0.01), and 40 cm (P < 0.05). ET_S was significantly correlated with soil moisture content at depths of 10 cm (P < 0.05) and 20 cm (P < 0.01). ET_R was significantly correlated with soil moisture content at depths of 10 and 20 cm (P < 0.01), and 40 cm (P < 0.05). Soil moisture content at depths of 10 and 20 cm was significantly correlated with ET_S+R (P < 0.01). The correlation between E, ET and soil water content at depths from 60 to 220 cm was not significant (P > 0.05).

The canopy height, canopy projection area, new shoot/branch length and LAI of S. passerina in the lysimeter #2 increased by 44, 126, 51 and 30%, respectively, from 2015 to 2016. The corresponding values for R. soongorica in lysimeter #3 were 40, 43, 50 and 37%, respectively. For the mixed community in the lysimeter #4, they were 18, 12, 34 and 15%, respectively (Table 1). The cumulative ET increased exponentially with increasing canopy height, canopy projection area, new shoot length and LAI (Figure 7).

Generally, ET exceeds precipitation inputs on both daily and annual time scales in desert ecosystems, where substantial bare soil evaporation with low plant cover drives the ecohydrology process (Noy-Meir 1973; Testi et al. 2004). In the present study, the ET/P ratio varied from 1.07 to 1.15 in 2015, and varied from 0.98 to 1.14 in 2016, suggesting no deep drainage. The E/P ratio was 0.95 in 2015 and 0.88 in 2016, suggesting only a tiny fraction of the rainfall contributed to the increase in soil water storage for the bare area. Actually, no percolation was observed for bare lysimeter and vegetated lysimeters. Wang et al. (2004c) suggested that the ET/P in natural vegetation of the Tennger Desert was 1.18, and the soil evaporation to precipitation ratio (E/P) in the shifting sand dune area was 0.71. Eddy covariance measurements showed that the ET/P was 1.03, 1.07, 0.82, and 1.11 from 2009 to 2012, respectively (Gao et al. 2016). Our findings also agree well with research on ET in other regions. In the Gurbantonggut desert in western China, the annual ET/P of the halophyte community was 1.20 in a dry year (ET was 155 mm, P was 129 mm) and 1.22 in a wet year (ET was 259 mm, P was 213 mm) (Liu et al. 2012). In the Chihuahua desert grassland, long-term measurements showed that ET/P was 1.09, with an average annual ET of 299 mm and a precipitation of 272 mm (Mielnick et al. 2005). As expected, the results of the actual ET shown in our study confirm the high reliability and accuracy of the large automatic weighing lysimeter as compared to previous studies in the arid desert region. It also highlights the importance of E, which is mainly dependent on precipitation and contributes to most of the actual ET in arid areas (Zhang et al. 2017).

A significant correlation between accumulative E, ET and P (R² > 0.9786, P < 0.01) indicates that both E and ET vary with P, which was consistent with the previous findings of Wang et al. (2004b). Precipitation in the study area is largely composed of small events with relatively short dry intervals, 64% precipitation events are less than 5 mm with 80% of dry intervals less than 10 days in length (Wang et al. 2005). Episodic large rainfall events (>10 mm) accounted for only 4 and 7% of the total precipitation event in 2015 and 2016, and the amount of large precipitation events accounted for 31 and 38% of the total precipitation in the growing season. Large precipitation events sharply increased E and ET on the daily time scale, especially E, then led to larger annual E and ET as our results indicate (Figures 2 and 3). Due to low canopy coverage, more incident radiation transmitted to the bare soil resulted in greater soil evaporation after large rainfall events (Zhang et al. 2010). On the other hand, they can elevate the soil moisture at 0–20 cm in a greater number of growing season days (Heisler-White et al. 2008). The occurrence of large precipitation events, moreover, are related to the distinguishing of wet and dry years, and likely led to increased soil moisture at greater depths (Sala et al. 1992). Our results indicated that E and ET were mainly affected by wind speed, followed by air temperature, solar radiation and relative humidity on a daily time scale during the growing season (Figure 5). Similar results have been reported in previous studies. For example, during 1955–2008 in northwest China, Huo et al. (2013) found that the contribution of wind speed to the decrease in reference evapotranspiration (ET₀) was greater than that of other meteorological variables. For ET₀, wind speed was the most sensitive meteorological variable, followed by relative humidity, temperature and radiation. Liu et al. (2010) reported that the meteorological variables for ET₀ varied with location. In addition, our results showed that the meteorological factors regulating E and ET are significantly rainfall event-dependent (Table 3). The results of the present study (Figures 4 and 5, and Table 3), coupled with those from literature, allow us to assume that wind speed is the dominant environmental factor affecting actual E and ET on a daily time scale, whereas precipitation controls patterns of actual E and ET on an annual time scale. It is notable that relative humidity is one of the driving meteorological factors for E and ET within 8 days after a large rainfall event. Therefore, the importance of the spatiotemporal scale and occurrence of rainfall events should be taken into account when analyzing the changes and influencing factors of E and ET.

Soil moisture is a crucial link between hydrologic and biogeochemical processes because of its direct influence in transpiration and indirect influence; for example, its dominant effect on nitrogen mineralization, as well as the key abiotic limiting factor in desert regions (Rodriguez-Iturbe 2000; Wang et al. 2017). Low soil water contents provide little buffer against climate extremes, which is one of the reasons why desert environments are particularly vulnerable to climate variability (Scanlon et al. 2005). ET dominates the fate of soil water content, since the percolation is usually a very small component of the water budget (less than 5%) (Wilcox et al. 2003). Our study suggested that the soil moisture content at 10 and 20 cm depths within four lysimeters are significantly correlated with E and ET (P < 0.01, P < 0.05), which throws light on the importance of these depths of soil water content on ET (Table 4). ET is primarily water limited in arid ecosystems, where soil moisture is especially important and controls the available water for ET (Chen et al. 2008). Limited soil water supply results in a decline in the global land ET (Jung et al. 2010). Soil moisture replenishment by infiltration and hydraulic redistribution and loss through soil evaporation, plant transpiration and deep drainage results in ET being the most interactive component of the water budget (Loik et al. 2004). We found that for a relatively smaller rainfall event of 24.8 mm the wetting front was limited to 40 cm depth (Figure 6), whereas for a consecutive series of large rainfall events with a cumulative rainfall of 42.5 mm, the wetting front could reach a depth of 90 cm, which was comparable to the previous study in the same region (Wang et al. 2011). In other words, a heavy rainfall (>24.8 mm) can replenish the soil water in the rooting zone of S. passerina, R. soongorica and their mixed community since their roots are mainly distributed at a depth of 0–40 cm. In the mobile dunes of the Horqin Sandy Land, a rainfall amount of 13.4 mm with rainfall intensity of 40.1 mm h⁻¹ could replenish soil water at 60 cm depth (Liu et al. 2015). The rainfall events with higher intensity will make a greater contribution to the soil moisture replenishment (Wang et al. 2008). In contrast, neither 27 mm nor 19 mm precipitation events resulted in recharge at 30 cm at a shrub-dominated station in the Chihuahuan Desert, likely due to runoff generation (Reynolds et al. 2000). Only large rain events (1 day with >25 mm or consecutive days each with >15 mm rainfall) generated a considerable wetting front at 15 cm and little change was observed at 30 cm of the soil profile (Scott et al. 2006). The presence of the vegetation allows precipitation to infiltrate into deeper soil layers (Figure 6), most probably because vegetation cover affects soil infiltration by providing more opportunities for infiltration (reducing runoff generation) and by modifying the structure of the soil pore spaces as a result of the formation of the root system (Huang et al. 2013). In addition, the presence of biological soil crusts with vegetation induced large soil moisture content beneath 10 cm (Wang et al. 2007), resulting in a relatively higher ET value.

Vegetation plays a dominant role in controlling the ET process by modifying transpiration (T) through stomata of plant leaves and E of intercepted precipitation in these regions, particularly in arid ecosystems (Méndez-Barroso et al. 2014). Shrubs were found to significantly reduce the soil evaporation under canopy in the Negev Desert (Kidron 2009). Conversely, the covering of vegetation hindered the rapid formation of a dry sand layer which is a constraining factor for soil water evaporation (Liu et al. 2015). The presence of biological soil crusts would not greatly change soil evaporation in arid and semiarid environments (Xiao et al. 2010). A contradictory result was obtained by Kidron & Tal (2012), who that found crusted plots have an average evaporation rate of 1.38 times higher than the bare sand and decreased evaporation under biocrusts. We found that single shrub communities have higher annual ET compared to associated shrub communities (Figure 2), because the transpiration rate of the associated shrub community was lower than that of the single shrub community (Su et al. 2012). In addition, our results indicated that lysimeter #2 exhibits a larger increase in canopy height, canopy projection area, new shoot length and LAI compared to two other vegetated lysimeters. This is mainly due to C₄ plant (S. passerina) which has more rapid growth than C₃ plant (R. soongorica) (Pearcy et al. 1981), while lysimeter #4 exhibits a small increase in morphological traits, most probably because the association retarded the growth of both S. passerina and R. soongorica (Su et al. 2012). LAI is the main vegetation characteristic that affects ET (Barbour et al. 2005), and as LAI increased, ET increased logarithmically (Zhao et al. 2016). However, our study indicated that new shoot length with a high coefficient of determination should be a more important contributing factor to a higher cumulative ET, as shown in Figure 7(c). Accordingly, we argue that the growth parameters such as canopy height, canopy projection area, new shoot/branch length and LAI explain the variation of xerophytic shrub ET at long term scales, for the reason that those growth parameters indirectly indicate an increase in vegetation biomass productivity associated with high transpiration (Scanlon et al. 2005; Suzuki et al. 2007).

We identified that annual scale variations in actual E and ET of typical xerophytic shrubs are mainly driven by precipitation and their diurnal dynamics are mainly caused by wind speed. However, diurnal variations of E and ET mainly depend on relative humidity after a large rainfall event within 8 days. Soil moisture within 40 cm of soil depth is significantly correlated with E and ET of typical xerophytic shrubs. The cumulative ET increased exponentially with increment of canopy height, canopy projection area and LAI, especially new shoot length. The cumulative ET of single S. passerina and R. soongorica community was greater than that of the associated shrub community in two growing seasons, indicating that single distributed S. passerina and R. soongorica communities will increase ET. Moreover, higher E/P suggested the relatively important contribution of soil evaporation to ET in the water-limited arid desert ecosystems. Our findings highlighted that research into ET trends for S. passerina and R. soongorica communities is particularly important in the desert region of northwest China. Furthermore, an in situ, long-term continuous study of ET by large weighing lysimeters can also lead to sufficient data for validation of ET model simulation.

This work was funded by the National Key Research and Development Program of China (2016YFE0203400), and the National Natural Science Foundation of China (Grant nos. 41771101 and 41530750).