How afforestation affects the water cycle in drylands: A process‐based comparative analysis

The world's largest afforestation programs implemented by China made a great contribution to the global “greening up.” These programs have received worldwide attention due to its contribution toward achieving the United Nations Sustainable Development Goals. However, emerging studies have suggested that these campaigns, when not properly implemented, resulted in unintended ecological and water security concerns at the regional scale. While mounting evidence shows that afforestation causes substantial reduction in water yield at the watershed scale, process‐based studies on how forest plantations alter the partitioning of rainwater and affect water balance components in natural vegetation are still lacking at the plot scale. This lack of science‐based data prevents a comprehensive understanding of forest‐related ecosystem services such as soil conservation and water supply under climate change. The present study represents the first “Paired Plot” study of the water balance of afforestation on the Loess Plateau. We investigate the effects of forest structure and environmental factors on the full water cycle in a typical multilayer plantation forest composed of black locust, one of the most popular tree species for plantations worldwide. We measure the ecohydrological components of a black locust versus natural grassland on adjacent sites. The startling finding of this study is that, contrary to the general belief, the understory—instead of the overstory—was the main water consumer in this plantation. Moreover, there is a strict physiological regulation of forest transpiration. In contrast to grassland, annual seepage under the forest was minor in years with an average rainfall. We conclude that global long‐term greening efforts in drylands require careful ecohydrologic evaluation so that green and blue water trade‐offs are properly addressed. This is especially important for reforestation‐based watershed land management, that aims at carbon sequestration in mitigating climate change while maintaining regional water security, to be effective on a large scale.


| INTRODUC TI ON
Some regions of the earth are becoming greener as a result of climate warming, CO 2 fertilization, and land-use change (Chen et al., 2019;Zhang, Song, Band, Sun, & Li, 2017). One of these "greening up" regions is northwestern China, where the Chinese Government implemented the world's largest afforestation programs (Chen et al., 2018;Liu, Li, Ouyang, Tam, & Chen, 2008). These programs are planned until 2050 to reduce soil erosion and land degradation, adapt to climate change, alleviate poverty, and improve and restore land quality as well as watershed ecosystem services (Zhang & Schwärzel, 2017). One hotspot region for these programs is the Loess Plateau, which supplies a large amount of water and sediment for the lower reaches of the Yellow River. The Loess Plateau is the size of France and comprises up to 300 m thick loess deposits, which are windblown silt-sized sediments. Soils developed from loess are fertile and easily cultivated but extremely prone to erosion. Centuries of land use and severe soil erosion have led to more than 70% of the former flat plateau becoming a gully-hill dominated landscape (Zhao, Mu, Wen, Wang, & Gao, 2013). Bryan et al. (2018) reviewed China's investment strategies for land-system sustainability and concluded that the large-scale afforestation programs were successful and set a world example to address future challenges; however, the afforestation had led to unintended local and regional water shortages. This is alarming for the health of Asia's third largest river, the Yellow

River.
Analyses on streamflow and land-use change as well as numerous modeling studies at the catchment scale (Feng et al., 2012;Sun et al., 2006;Zhang et al., 2014) show that land-use change plays a bigger role for observed water yield reductions than does climate change. Although a growing number of plotscale studies quantify soil water storage alterations due to afforestation (Jia, Shao, Zhu, & Luo, 2017;Liu et al., 2018;Yang, Wei, Chen, & Mo, 2012), none address the interplay of rainfall, evapotranspiration, soil water dynamics, and seepage of dryland forest plantations. There are no rigorous process-based comparative measurement studies on how forest plantations affect the water balance components and alter partitioning of rainwater into green (evapotranspiration) and blue water flows (surface and subsurface). Wang et al. (2012) pointed out that the relationship between woody plants, understory, and soil water is barely understood, and the degree to which one offsets another relies on the scale at which these impacts are quantified. Newman et al. (2006) also argued that our understanding of feedback and interactions between biotic and hydrologic components of environmental systems depends mainly on ecosystem process models, and a lack of data has restricted their application. For instance, little is known about the partitioning of evaporative fluxes in dryland afforestation (Gimeno, McVicar, O'Grady, Tissue, & Ellsworth, 2018) when this would provide a deeper understanding of the biological regulation of the hydrological cycle and the impact of soil water dynamics on vegetation performance (Huxman et al., 2005;Kool et al., 2014;Newman et al., 2006). Black locust (Robinia pseudoacacia L.) is a major tree species for plantations on the Loess Plateau. Its good growth attributes have made it one of the most commonly cultivated deciduous tree species worldwide. It is native to North America and has been introduced to many different regions for wood, fuel, forage, and beekeeping.
Black locust rivals poplar as the second most planted broadleaved tree species worldwide, after the eucalypts (Redei et al., 2018).
Black locust was introduced in China early in the 20th century and is now widely used for afforestation as a pioneering tree species (Wang, 1992). Currently, it represents >90% of plantation forests on China's Loess Plateau . However, water balance studies on black locust plantations are rare, and moreover, published water consumption data for this species are often biased. In earlier works (Podlasly & Schwärzel, 2013;Schwärzel, Zhang, Strecker, & Podlasly, 2018), we developed a specific calibration method to improve the estimation of the water consumption of black locust and implemented field experiments that partitioned evapotranspiration of black locust stands with understory.
Consequently, here we focus on (a) how the overstory and understory contribute to the total water consumption of black locust plantations, (b) which environmental factors control the various evaporative fluxes, and (c) if and to what extent black locust plantations alter the water cycle. To address these points, we use established ecohydrological measurements of black locust plantations versus natural grassland (as a reference system) at adjacent sites under the same soil, topography, and climate conditions . Our work will contribute to addressing data gaps on feedback and interactions between biotic and hydrologic components in water-limited environments.

| Weather stations
An open-land weather station was installed in 2012, ~300 m east of the black locust plantation. Global and net radiation (Kipp & Zonen), air temperature, relative humidity, wind direction and speed (Thies), and soil temperature at the soil surface and different depths were recorded in 15 min intervals. Rainfall was measured using a tipping bucket system with heating (Lambrecht). Air and soil temperature, relative humidity, and global and net radiation were also measured under the black locust canopy. Two stainless steel throughfall troughs (5 m long, 0.16 m wide) draining into a tipping bucket were installed beneath the canopy and operated from the end of March until mid-October.

| Soil water conditions
To measure soil water dynamics, a wireless sensor network (FZ Jülich) was installed under the black locust canopy and on the grassland. Soil moisture was measured using a spade sensor (sceme.de GmbH) with a measurement resolution of 0.01 (L 3 /L 3 ; Qu, Bogena, Huisman, & Vereecken, 2013

| Understory and grassland evapotranspiration
To quantify water consumption of understory and grass, two weighable field lysimeters were installed: one under the black locust canopy and one on the grassland. The lysimeter comprised a polyethylene container (2.50 × 2.30 × 1.50 m); a stainless steel lysimeter vessel (surface area 1.0 m 2 , length 1.70 m) filled with undisturbed soil; a weighing system; and a unit to automatically regulate soil moisture and temperature at the lower bottom of the lysimeter. (For details of lysimeter installation and automated control of its lower boundary, see Podlasly and Schwärzel, 2013.) Decreases in lysimeter weight (logged every 30 min) were caused by root water uptake and soil evaporation. Thus, we use understory evapotranspiration (ET us ) and grassland evapotranspiration (ET grass ) when referring to water consumption quantified by the lysimeter, calculated as follows: where P net is net precipitation (measured by throughfall troughs), P is open-land precipitation (L/T), D is drainage water collected at the lysimeter bottom (L/T), ΔM is change of lysimeter mass (1 M/T = 1 L 3 /T), ρ w is water density (M/L 3 ), r is lysimeter radius (L), and ΔT is the amount of water added to control pressure head at the lower bottom of the lysimeter (L/T).

| Overstory transpiration
Granier-style sap flow sensors were used to quantify stand contribution to total evapotranspiration. Probes were 20 mm long with diameter 1.5 mm (Ecomatik). Fourteen trees were equipped with sap flow sensors. (For installation details, see Schwärzel et al., 2018.) Temperature differences between the heated and unheated needles were measured in 60 s intervals and data logged as 60 min averages.
Hourly sap flow densities, F d (L 3 /L 2 ·T), were calculated from observed temperature differences using Equation (3) (Granier, 1985): where ΔT (K) is measured temperature difference between the two needles, ΔT M (K) is ΔT value at F d = 0, and coefficient a and exponent b are fitting parameters. Instead of Granier's original parameters, tree-specific parameters derived from in situ calibration on a living tree at our experimental plot (Schwärzel et al., 2018) were used to convert observed temperature differences into sap flow densities. Additionally, we used the soil water balance method to test the reliability of the tree-specific calibration. The coefficient a obtained from the fit was 3.29 kg m −2 s −1 , and the parameter b was very similar to the 1.231 reported by Granier (1985). To consider potential nocturnal fluxes due to both transpiration and recharge, ΔT M was estimated according to Lu, Urban, and Zhao (2004) and determined using BaseLiner software (Schwärzel et al., 2018). No functional relationship between the daily total sap flow density and corresponding DBH was found. Thus, daily sap flow densities of all sample trees were averaged to estimate daily transpiration (T BL ) of the black locust stand (mm/day) using averaged flux densities (J avg ), To determine conducting sapwood, 10 trees-with DBH similar to trees used for sap flow measurement-were harvested and (1) ET us = P net − D −ΔM∕ w r 2 +ΔT, corresponding tree-ring widths of the outermost rings were measured using stem disk at breast height. Regression between DBH and corresponding sapwood area was used to calculate total sapwood area of the plot using the frequency of tree DBH. The tree-cutting technique was applied to visualize water-conduction pathways in living black locust trees (Schwärzel et al., 2018).

| Seepage
We applied three different methods to estimate seepage: (a) lysimeter method, (b) zero-flux-plane (ZFP) method, and (c) calculation of annual seepage rate as residual of the field water balance in the surface soil zone of the grassland and forestland site at a level position (i.e., no surface runoff). The latter can be described as: where S is seepage/deep percolation below the soil zone, P gross is precipitation on open land, P tf is throughfall, I is canopy interception of rainfall, and ΔW is water storage change in the soil profile. (5) and (6)  seepage. For shorter time intervals, soil water storage changes must be considered when estimating seepage under forest and grassland; additionally, evapotranspiration and drainage can occur simultaneously. Consequently, the ZFP method was applied using the soil water content and soil pressure head measurements as a function of time and depth as described above. ZFP is defined as a plane separating two zones of upward and downward movement of water in unsaturated soils with simultaneous evapotranspiration and seepage (Khalil, Sakai, Mizoguchi, & Miyazaki, 2003).

Terms in Equations
If ZFP is located below the root zone, soil moisture changes above the ZFP are due to evaporation from soil and root water uptake, whereas soil moisture changes below the ZFP are caused by seepage. A limitation of the ZFP method is preferential flow during storm events, implying that much of the soil above the ZFP is bypassed whereas soil below the ZFP is wetted (Cooper, Gardner, & Mackenzie, 1990;Schwärzel et al., 2009). Under such circumstances, this method would overestimate evapotranspiration and underestimate seepage (Cooper et al., 1990). At our site, preferential flow was important in the recharge of plant-available soil water and seepage, as indicated by a rapid decline in soil pressure head at depth of 80 cm in September 2012, July 2013, and September 2013 ( Figure 1). Consequently, ΔW in Equations (5) and (6) was not determined on a daily timescale but calculated-as the difference in soil water storage at the beginning and end of the balance period-over the entire period in which a ZFP existed and the hydraulic gradient below the main root zone indicated (5) S grass = P gross − ET grass −ΔW,

Rain
Air temperature at 2m height (daily average) (a) downward movement of water. The daily values of rainfall and evaporation components of Equations (5) and (6) were summed for these balance periods.

| Data gap filling and further calculations
Due to occasional instrument failure and unfavorable weather, data gap-filling was needed. Meteorological data were obtained from the Changwu Agro-Ecological Experimental Station (30 km from our field) and regression analyses applied to calculate missing data.
In a very few cases, the canopy or grassland lysimeter delivered no or implausible data and these gaps were filled by regressions using data from the other lysimeter. Missing overstory transpiration data were filled by hydrological modeling using BROOK90 (Federer, Vörösmarty, & Fekete, 2003), which estimates daily in-

| Weather and soil water dynamics during the study
Atmospheric evaporative demand during the main growing season (May-September)-characterized by grass reference evapotranspiration-was significantly lower (770 mm) in 2012 and slightly lower (810 mm) in 2013 than the long-term average (840 mm). This was partly due to higher seasonal rainfall (460 mm in 2012 and 630 mm in 2013) than the long-term average (420 mm). Figure 1a shows a distinct temporal pattern of rainfall.
There was almost no rainfall between the end of September and the beginning of April. In contrast, 60%-80% of annual rainfall occurred during June-September mostly in heavy falls ( Figure 1a). however, VPD at the forest floor remained high as shown by small differences in VPD between forest floor and grassland ( Figure 2c).
This indicates a significant degree of coupling through the forest canopy layer.

| Diurnal and seasonal variations of evaporative fluxes
There was a stark contrast in evaporation among overstory, understory, and grass as shown for the two hot sunny days without rainfall before (16 June) and after (14 July) the onset of the rainy season in 2012 ( Figure 3). On these days, the canopy was fully developed and meteorological conditions-diurnal courses of global radiation, air temperature, and VPD and daily averages of weather variables-were comparable (Figure 3a,b and Table 2). However, soil water conditions for these 2 days differed significantly (Figure 3c Overstory conductance showed little sensitivity to changes in soil moisture (Figure 3g,h). Contrary to the latter, evaporation from grassland responded to increased soil water availability (characterized by pF-values): under dry soil conditions (16 June), the ratio between daily evaporation from grassland to daily grass reference evapotranspiration was 67% (Table 1), but for wetter soil (14 July), the ratio was 87%. For understory and overstory evaporation, the ratios between measured evaporation and calculated grass reference evapotranspiration remained unchanged under changing soil water conditions (Table 1). Compared to overstory and grassland, understory conductance often had a multipeaked daily cycle with maxima before noon and late afternoon (Figure 3g,h). Furthermore, conductance varied much more throughout the day for understory and overstory than grassland. The coefficient of variation of canopy conductance was 50% for overstory and understory, and <20% for grassland. April-October 2012. Due to canopy shading, daily evaporation rates were usually lower for understory than grassland; however, the differences were small (Figure 4a,b). Furthermore, daily values of overstory transpiration (Figure 4b) were markedly lower than grassland ( Figure 4a) and understory evaporation (Figure 4b).
The highest fluxes were during June-August with daily maxima of 4.7 mm for grassland evaporation, 3.7 mm for understory evaporation, and 1.8 mm for overstory transpiration. During June-August, daily mean evaporation from grassland, understory, and overstory was 2.5, 1.8, and 1.1 mm, respectively. Low daily evaporation rates were generally associated with rainfall. Daily rates of overstory transpiration varied considerably less than evaporation from grassland or understory (Figure 4b). During June-August, the coefficient of variation of daily overstory transpiration was 30%, compared to 48% for grassland and 50% for understory. These differences among vegetation covers could be caused by different mechanisms of evaporation control, as discussed in the next section.

| Influence of environmental conditions on evaporation
The relationship between daytime data of net radiation and daily evaporation rates of the contrasting vegetation covers was plotted for June-August 2012 and 2013 when the canopy was fully developed ( Figure 5). Clearly, overstory transpiration was a maximum at net radiation of 15 MJ m −2 day −1 (Figure 5a), whereas evaporation from grassland ( Figure 5b) and understory (Figure 5c) was largely controlled by radiation. However, compared to grassland, understory data were much more scattered. In general, differences in relationships between radiation and evaporation for the contrasting vegetation covers can be attributed to differences in stomatal control of evaporation. For a better understanding of the latter, surface Note: Hourly data are presented in Figure 3.
conductance of grassland, understory, and overstory was estimated on an hourly basis as previously outlined. There was a moderate relationship of understory conductance with VPD (R 2 = .50, not shown) and with soil pressure head (R 2 = .48, not shown; Figure 7a,b). This indicated that daily understory evaporation was not just controlled by radiation as discussed above ( Figure 4) but also by VPD and soil moisture conditions. In contrast to understory, only a weak relationship between VPD and grassland conductance (R 2 = .10, not shown) was observed (Figure 7c), but Figure 7d revealed a trend for grassland conductance to decrease markedly when topsoil under grassland dried up (pF > 3.0, R 2 = .19, not shown). It was discussed above that grassland transpiration was largely controlled by radiation (Figure 4). However, Figure 7d indicates that soil water conditions had strong control over the evaporation flux from grassland if a certain soil water deficit was reached (characterized by the soil pressure head of topsoil). This was discussed above concerning the diurnal dynamics of evaporative fluxes ( Figure 3).

| Field water balance of contrasting vegetation covers
The water balance components for each month are shown in Table 2.
The lack or presence of leaves of the forest canopy controls the partitioning of evaporation as well as of rainfall. During the leafless period from the end of October to mid-April, differences in evaporation between the contrasting vegetation covers were relatively small. Large discrepancies in evaporation between the contrasting vegetation covers were observed as soon as the canopy was fully developed in May. This was due to the combination of overstory transpiration, evaporation of rainwater intercepted by the forest canopy, and evaporation from the understory.   (Table 2). This lag is because drainage water collected at a depth of 1.75 whereas the ZFP method balances soil water storage changes over a depth of 1 m.
Partitioning of annual rainfall in grassland and forestland showed that black locust plantation lost 92% of annual rainfall as ETI, and grassland lost 80% (Figure 9). Under the forest, most water was taken up by the understory, representing 58% of annual rainfall compared to only 26% by tree roots. Canopy interception was low (8%) due to frequent rainstorm events. Annual seepage-calculated as residual F I G U R E 7 Relationship (a, c) between daily means of canopy conductance (g c ) and vapor pressure deficit, VPD, and (b, d) between daily means of g c and the logarithm of soil pressure head (log |hsoil|) at depth of 20 cm for understory (a, b) and grassland (c, d). Shown are daily data for days with rainfall < 1 mm during 2012 and 2013 growing seasons  F I G U R E 8 Seasonal changes in hydraulic gradient below the main root zone at depth of 70 cm under forest and grass, daily rainfall, and the sum of seasonal seepage (negative numbers) under grass and forest estimated using Equations (5) and (6)

| Environmental control of evaporation
The black locust canopy was well coupled to the atmosphere as the strong relationship between VPD and surface conductance demonstrated ( Figure 6). Moreover, the omega factor, which describes the degree of decoupling between transpiration and the saturation deficit in the boundary layer, was on average 0.24 during the main growing seasons (data not shown), indicating that black locust transpiration was strictly physiologically regulated.
The results suggest a conservative water use strategy of black locust in avoiding turgor loss and xylem cavitation-supported by the small sensitivity of its canopy conductance and transpiration to rainfall-induced changes in soil moisture conditions (Figure 3). and 28 year old trees after rainfall, respectively. Du et al. (2011) showed that rainfall-induced increases in soil moisture were accompanied by increases in normalized sap flux densities; for instance, maximum flux density was ~0.57 and ~0.75 on a pre-and post-rainfall day, respectively. However, Du et al. (2011) acknowledged that the differences in VPD and solar radiation (both considerably larger on the post-rainfall day) between the 2 days made it difficult to determine the effect of rainfall on diurnal patterns.
These discrepancies might be attributed to differences in available soil water during the measurement periods. As outlined above, the study area experienced above average rainfall. Due to frequent heavy rainfall during both growing seasons, black locust showed no severe water shortage (Figure 1). Thus, no responses of transpiration to rain pulses were detected. However, it is possible that during seasons of below average rainfall soil under black locust dries more intensively and significant responses of transpiration to rainfall-induced increase in plant-available soil water might occur.
Deviating from the general expectation that grass and other short vegetation is decoupled from the atmosphere (Jarvis & Mcnaughton, 1986), our study showed that grassland was moderately coupled to the atmosphere as demonstrated by the analysis of the relationship between evaporation from grassland and net radiation and between surface conductance and VPD ( Figures 5 and 7). Moreover, the omega factor was on average 0.47 during the main growing seasons (not shown) indicating moderate coupling between the grass surface and atmosphere. As shown above (Figure 1 FLUXNET data. They found for grasses that the degree of coupling increased significantly with decreasing 3 month rainfall. Compared to grassland, our understory seemed better coupled to the atmosphere as the relationships of evaporation and net radiation with conductance, VPD, and topsoil moisture conditions revealed ( Figures 5 and 7). A further indication of the considerable coupling between the understory and atmosphere was the small difference in VPD between forest floor and grassland ( Figure 2). However, VPD was not the main driver of understory evaporation-rather, it was radiation-but evaporation rates were controlled by canopy conductance, VPD, and soil moisture. This is discussed in more detail in supplementary material where measured and simulated ET us are compared. The relationship between environmental factors and understory canopy conductance should be considered when modeling evapotranspiration of multilayer forest stands.

| Reliability of measured water cycle components
The reliability of our sap flow measurements is discussed under data gap filling procedures (supplementary material) and in Schwärzel et al. (2018). Moreover, Ma et al. (2017) and Schwärzel et al. (2018) argued that most previous experimenters significantly underestimated the sap flow of black locust using Granier's universal calibration equation-Equation (3). Investigators (Jian, Zhao, Fang, & Yu, 2015;Ma et al., 2017;Zhang et al., 2018) who modified the Granier method found that black locust transpiration rates were consistent with the generally accepted notion that afforestation resulted in a significant discharge reduction in the Loess Plateau through elevated evapotranspiration (Schwärzel et al., 2018). Ma's and Zhang's studies were comparable to our study in terms of maximum LAI (2.4-2.8) and mean annual rainfall (580-592 mm) whereas mean annual rainfall was only 420 mm for Jian et al. (2015). The latter quantified stand transpiration of a 28 year old black locust forest (mean LAI ~ 2.5) over a period of 5 years; their average daily transpiration was 1.1 mm, which agrees well with our findings. Ma et al. (2017) reported average daily transpiration values of 2.1 mm for 2015 and  (Liu et al., 2018;Nan, Wang, Jiao, Zhu, & Sun, 2019;Scott & Prinsloo, 2009;Wang et al., 2012).
Understory evaporation was a significant component of the water balance of the studied plantation (Table 2, Figure 8). Similar findings were reported from other forests. Understory vegetation of Mediterranean oak savannas (Dubbert et al., 2014;Paço et al., 2009) and eastern Siberian larch forests can contribute up to 50% of total evaporation (Iida et al., 2009;Kelliher et al., 1997). About 20%-30% of net canopy mass and energy exchange occurred at the floor of boreal pine forest (Baldocchi & Vogel, 1996). Gimeno et al. (2018) reported for mature native eucalyptus woodland in a humid temperate-subtropical transitional climate that the understory con- This was lower than our average result of 1.8 mm, but the differences are reasonable considering that our lysimeter simultaneously captured root water uptake and evaporation from soil. Jian et al. (2015) found that soil evaporation accounted for >40% of total seasonal (May-September) forest ETI. Jiao et al. (2018) reported that soil evaporation represented 55%-59% of total seasonal forest ETI. Wang (1992) used water balance to estimate evapotranspiration of a 14 year old black locust stand, and reported annual evapotranspiration of 630 mm for a year with rainfall of 700 mm, consistent with our findings (Table 2).
Comparing seepage estimates using different methods is another means of evaluating the reliability of our measurements.
The drainage water collected at the bottom of the lysimeter under forest was 69 mm in 2013 (Table 2), and the corresponding seepage using the ZFP method and calculation as residual of annual water balance were 63 and 56 mm, respectively (Figures 8 and 9).
The drainage water from the bottom of the lysimeter under grass was 178 mm in 2013 (Table 2), while the seepage estimate using ZFP was 188 mm (Figure 8) and that calculated using annual water balance was 140 mm (Figure 9). The discrepancy (~20%) between drainage water collected from lysimeters and those calculated using the water balance indicates only small changes in annual soil water storage at our site. The soil pressure head values at a depth of 80 cm were slightly higher at the beginning than at the end of 2013 (Figure 1). However, good agreement between the three seepage estimates indicates that our measurement results are plausible.

| Dilemma of soil conservation through afforestation
Our field observations at adjacent plantation and grassland sites showed that afforested land removed significantly more water from soil to atmosphere than grassland. Afforestation promotes formation of multilayer stands with higher LAI and deeper root systems, thus increasing evapotranspiration (Table 2; Figure 4).
Contrary to common belief, the understory, not the overstory, was the main water consumer in this plantation. The enhanced evapotranspiration of afforested land was at the expense of seepage (Figures 8 and 9; and Table 2). Our measurements indicate that significant seepage under forests might only occur in years with rainfall exceeding the annual average of 590 mm. Moreover, seepage flow below the root zone only occurred during the rainy season of June-September ( Figure 8). Our data suggest that preferential flow through the root zone is an important mechanism of seepage formation in the study area. In a study using the chloride mass balance method for recharge estimation under different land covers, Gates et al. (2011) and Huang and Pang (2011) found chloride accumulation under tree and shrub plantations, but not under agricultural land-use, suggesting that afforestation prevents seepage-a conclusion confirmed by our study. This decrease in seepage can be explained by vertical niche separation between trees and grasses. A comparison of soil moisture dynamics between grassland and forest-indicated by the seasonal course of soil pressure head as a function of depth-demonstrated that the understory sourced water from topsoil, while the overstory mainly took water from subsoil (Figure 1). Such a two-source system was proposed by Walter (1939) to explain the equilibrium of trees and grasses in savannas. This concept was later simplified by Walker and Noy-Meir (1982) and described by Ward, Wiegand, and Getzin (2013) as follows-grasses with their intense and shallow root system use water only from subsurface layers while woody savanna trees use little of the topsoil water but would have exclusive access to and primarily rely on subsoil water below the grass roots, consistent with our study. Due to the vertical niche separation, the soil profile under the forest with understory was much more depleted during the season than under grassland. Thus, more rainwater was needed to recharge plant-available soil water before seepage could occur.
While effective soil erosion control by afforestation can lower surface runoff and "green up" China's drylands, these new forests have resulted in unintended local and regional water shortages, particularly in the Loess Plateau (Wang et al., 2012;Yao, Xiao, Shen, Wang, & Jiao, 2016;Yuan et al., 2018;Zhang, Podlasly, Feger, Wang, & Schwärzel, 2015;Zhang et al., 2014;Zhao, Mu, Strehmel, & Tian, 2014). Rainfall is the primary source of water for plant growth there and the amount of rainfall infiltrating into the soil depends on soil surface conditions; they control the partitioning of rainfall into overland flow and subsurface flow feeding soil water storage (Falkenmark & Rockström, 2006).However, water is not only partitioned at the soil surface but also in the root zone. At the partitioning point in the root zone, water is divided into seepage which eventually contributes to groundwater recharge (blue flow) and into evaporation (green flow; Falkenmark & Rockström, 2006). Compared to grassland, green flow from afforested land is enhanced by the vertical niche separation between overstory and understory. The dilemma of soil erosion control in drylands is clarified in Figure 9. Establishment of vegetation as a measure to reduce soil erosion significantly alters the portion of blue and green flows in water-limited regions. The relative contribution of the different green flow components to total evapotranspiration depends on forest stand structure (e.g., density, LAI, and species composition). Any management-induced changes in this structure (e.g., thinning) thus alter the partitioning of evaporative fluxes and will also lead to a complex change in seepage and water yield (Ganatsios, Tsioras, & Pavlidis, 2010;Jackson et al., 2005;Xuan Dung, Miyata, & Gomi, 2011). To ensure adequate levels of water supply, at both local and regional scales, forest evapotranspiration requires better control by vegetation management (Zhang & Schwärzel, 2017). Our results reveal that ET us of the studied forest needs to be lowered to promote seepage. To increase infiltration and seepage while minimizing water consumption by vegetation for effective forest management, introducing tree species that generate stem flow may increase seepage rates while the understory will be partly overshadowed due to the increase of LAI and so would reduce understory water consumption. Schwärzel, Ebermann, and Schalling (2012) showed that, for beech trees, crown and root architecture control stemflow formation, stem-flow infiltration, and subsurface transport of stem-flow through soil. As mentioned earlier, 60%-80% of seasonal rainfall in the Loess Plateau is storm events. It is well known that significant stem-flow formation occurs during such events (e.g., Levia & Germer, 2015). Introducing stem-flow-generating tree species (e.g., Chinese walnut) into black locust monocultures may help reduce soil erosion and ET us . Simultaneously, enhanced stem-flow formation could favor increased seepage rates and thus water yield. Such stem-flow-generating tree species could act as a biological rainwater harvesting system for increasing blue water flows. However, to our knowledge, such processes have never been studied in the Loess Plateau.

| Implications of global reforestation on water cycle
Forest restoration and tree-planting projects must be evaluated against the sustainable use of water resources and related ecological and socioeconomic consequences. From the Great Green Wall initiative in Africa's Sahel region to the Restoration Initiative in Asia, tree-planting is a popular strategy to mitigate climate change (Bastin et al., 2019;Griscom et al., 2017), land desertification and degradation, biodiversity loss, and food deficit. Such programs require large public and private investments, yet their sustainability is often not fully evaluated. The trade-offs between water loss and carbon sequestration have not been acknowledged in many regions (Feng et al., 2016;Jackson et al., 2005). Field research is expensive and hydrologic responses to tree planting takes time to observe. To fully understand how plantation forests alter the water cycle, important hydrological processes, such as seepage and soil and canopy evaporation, need to be fully considered. Where the partition of evapotranspiration is to be understood, its values for both the overstory and understory must be measured at the same time. This study is a pioneering attempt to quantify the impacts of dryland forestland versus grassland on the water cycle and compare their water fluxes addressing this knowledge gap. Our work showed that plantations in an arid/semiarid climate consume an amount of water close to (in normal rainfall years) and likely to exceed (in drier years) the annual precipitation. Planted forests deplete soil water and prevent deep seepage for groundwater recharge, thus, possibly further intensifying local and regional water scarcity and cause conflicts over water supply. This study provides local and national policymakers, NGOs, and UN organizations with evidence that good intentions to implement large-scale tree-planting in drylands can have drawbacks and unintended consequences to local and regional water security. With climate change resulting in more frequent and longer droughts over larger areas, a wellrounded evaluation of the effects of dryland tree-planting on the water cycle and of hydrological feedback is particularly critical and meaningful to avoid high costs (i.e., trade-off of ecosystem services) and loss of valuable investment.

ACK N OWLED G EM ENTS
This study was funded by the Deutsche Forschungsgemeinschaft (DFG Schw1448-3/1). We thank Atiqah Fairuz Salleh for final language editing.