Recent trends in vegetation greenness in China significantly altered annual evapotranspiration and water yield

There has been growing evidence that vegetation greenness has been increasing in many parts of the northern middle and high latitudes including China during the last three to four decades. However, the effects of increasing vegetation greenness particularly afforestation on the hydrological cycle have been controversial. We used a process-based ecosystem model and a satellite-derived leaf area index (LAI) dataset to examine how the changes in vegetation greenness affected annual evapotranspiration (ET) and water yield for China over the period from 2000 to 2014. Significant trends in vegetation greenness were observed in 26.1% of China’s land area. We used two model simulations driven with original and detrended LAI, respectively, to assess the effects of vegetation ‘greening’ and ‘browning’ on terrestrial ET and water yield. On a per-pixel basis, vegetation greening increased annual ET and decreased water yield, while vegetation browning reduced ET and increased water yield. At the large river basin and national scales, the greening trends also had positive effects on annual ET and had negative effects on water yield. Our results showed that the effects of the changes in vegetation greenness on the hydrological cycle varied with spatial scale. Afforestation efforts perhaps should focus on southern China with larger water supply given the water crisis in northern China and the negative effects of vegetation greening on water yield. Future studies on the effects of the greenness changes on the hydrological cycle are needed to account for the feedbacks to the climate.


Introduction
There has been growing evidence that vegetation greenness has been increasing in many parts of the northern middle and high latitudes during the last three to four decades (Xiao and Moody 2005, de Jong et al 2011, Guay et al 2014. The evidence is mainly from the satellite-derived normalized difference vegetation index (NDVI) or leaf area index (LAI). The 'greening' of the land surface can be attributed to various factors such as plant growth, afforestation/reforestation, and improved agricultural practices. Changes in vegetation greenness can alter the water balance by regulating matter and energy cycles (Bonan 2008).
Changes in vegetation cover or productivity influence the hydrological cycle mainly through modulating the processes of canopy interception, evaporation and transpiration (referred to as evapotranspiration or ET hereafter), and infiltration. The relationship between vegetation cover particularly forest cover and water yield remains hotly contested (Ellison et al 2012). Some studies showed that afforestation or reforestation could reduce available water supply (Zhang et al 2001, Andréassian 2004, Brown et al 2005, Jackson et al 2005. Increasing tree cover increases canopy interception and ET and thereby reduces runoff, leading to declining water availability. These studies are largely based on small-scale studies including paired-catchment experiments in small catchments. By contrast, some other studies showed that increasing forest cover can facilitate large-scale transport of water vapor and promote precipitation at regional to global scales, and therefore forest cover may have positive impacts on the hydrological cycle (Makarieva et al 2006, Liu et al 2008b, Sheil and Murdiyarso 2009, Jiang and Liang 2013. Despite the large number of studies, the effects of vegetation greening particularly afforestation on the hydrological cycle have been controversial. Trends of vegetation greening have been observed in many areas of China (Xiao and Moody 2004, Park and Sohn 2010, Li et al 2012, Chen et al 2014). China has the largest area of forest plantations in the world. Afforestation and reforestation have been nationwide efforts in China since the 1950s. To mitigate environmental degradation, the Chinese government also implemented multiple large-scale ecosystem restoration programs, including the 'Three-North Forest Shelterbelt Program', and the 'Natural Forest Conservation Program', and the 'Grain for Green Program'. These efforts substantially increased the national forest area (Gao et al 2014, Chen et al 2015, Zhang et al 2015b. Improved agricultural practices including the use of irrigation and chemical fertilizers and the substitution of high-yield crops for low-yield corps led to the increase of agricultural productivity Moody 2004, Xiao et al 2015). Global change factors including climatic warming, rising atmospheric carbon dioxide (CO 2 ) concentrations, and nitrogen deposition enhanced plant growth (Tian et al 2011, Xiao et al 2015. In the meanwhile, natural vegetation decreased because of ecosystem degradation and rapid urbanization , Liu et al 2012a, leading to decreases in vegetation cover or productivity ('browning') in some parts of the country. The changes in vegetation greenness could alter the regional carbon and water cycles and the local surface climate.
Several studies showed that increasing vegetation cover resulting from afforestation and ecosystem restoration in China reduced streamflow (Huang et al 2003, Bao et al 2012, Li et al 2014. For example, the simulations of a hydrological model indicated that forestation may reduce average water yield by 50% in the semi-arid Loess Plateau region in northern China and 30% in the tropical southern region (Sun et al 2006). By contrast, Xie et al (2015) indicated that afforestation practices had negligible effects on the hydrological cycle over the Three-North region of China. Moreover, the results based on an ecosystem model indicated that deforestation across China increased ET and decreased water yield while reforestation dcecreased ET (Liu et al 2008a). Previous studies therefore are inconsistent regarding the effects of afforestation/reforestation on the hydrological cycle. A recent study showed that the increase in urban areas and the reduction of rice paddy fields reduced ET by 23% and increased streamflow by 58% in the Qinhuai River basin in southern China (Hao et al 2015b). Better understanding how the changes in vegetation greenness affect ET and water yield has ecohydrological implications and can inform regional and national afforestation/reforestation policies.
In this study, we used a process-based ecosystem model and a satellite-derived LAI dataset to examine the effects of vegetation greening and browning on annual ET and water yield for China over the period from 2000 to 2014. First, the LAI dataset was used to assess the linear trends of vegetation greenness. Second, the 'greening' or 'browning' trend was removed on a per-pixel basis by detrending LAI. Third, we used a process-based ecosystem model-Boreal ecosystem production simulator (BEPS) along with the original and detrended LAI datasets to conduct two model simulations: one with original LAI (i.e., with trends of vegetation greening and browning) and one with detrended LAI (i.e., with no trends of vegetation greening or browning). Finally, the two model simulations were used to assess the effects of vegetation greening and browning on terrestrial ET and water yield.

Model description
We used the process-based ecosystem model-BEPS (Liu et al 1997) to assess how the trends in vegetation greenness affect annual ET and water yield in China. The BEPS model is run at half-hourly to daily time step, and is driven by meteorological variables (temperature, precipitation, incoming solar radiation, and relative humidity), remotely sensed LAI, land cover type, atmospheric CO 2 concentrations, and soil properties data. The BEPS model consists of hydrology, photosynthesis, energy balance, and soil biogeochemistry modules (Ju et al 2010b). This model stratifies a canopy into sunlit and shaded leaves, for which daily carbon fixation and transpiration are separately calculated (Chen et al 1999). ET of an ecosystem is calculated as the sum of canopy transpiration from sunlit and shaded leaves, and evaporation from soil surface and intercepted water by leaf surface. Both transpiration and evaporation are calculated using the Penman-Monteith equation (Chen et al 2005). The BEPS model has been continuously improved and has been applied to simulate hydrological processes for a wide variety of terrestrial ecosystems (Ju et al 2010a, Zhang et al 2013, Liu et al 2013b. More details about this model can be found elsewhere (Liu et al 1997, 2013b, Chen et al 1999, 2005, Ju et al 2006.

Driving datasets
In this study, vegetation greenness in China was characterized using a gridded LAI dataset derived from the moderate resolution imaging spectroradiometer (MODIS) (Liu et al 2015b). This LAI dataset consists of 8 day LAI composites with 500 m spatial resolution from 2000 to 2014. The LAI data were retrieved from the 8 day MODIS reflectance product (MOD09A1 V05) (Vermote and Vermeulen 1999) and the yearly MODIS land cover datasets (MCD12Q1 V051) (Friedl et al 2010) using a 4-scale geometric optical model (Deng et al 2006, Liu et al 2012b. Previous studies indicated that the 4-scale geometric optical model was superior to the algorithm used to produce the MODIS LAI product (Pisek et al 2007, Garrigues et al 2008. We used the 500 m resolution LAI dataset to examine the trends of vegetation greenness and to prescribe LAI for the BEPS model simulations. We used gridded daily precipitation, maximum and minimum air temperatures, incoming solar radiation, and relative humidity from a gridded meteorological dataset with 500 m spatial resolution. This dataset was interpolated from observations from 753 weather stations across China using the inverse distance weight method (Liu et al 2015b). The accuracy of the daily gridded meteorological dataset was evaluated at the site level using daily temperature, precipitation, and radiation measurements from 8 eddy covariance (EC) flux sites (http://chinaflux.org). The comparison showed that the daily radiation and temperature exhibited excellent agreement between the gridded dataset and the tower observations with R 2 values ranging from 0.77 to 0.92 for daily radiation and 0.90 to 0.99 for daily temperature, respectively (figures S1 and S2). The consistence was lower for daily precipitation (figure S3) because of the complexity of precipitation dynamics and the missing data of the EC sites. We also evaluated the gridded meteorological dataset using monthly 0.5°precipitation and temperature data in 2000 obtained from China Meteorological Administration (http://cdc.cma.gov. cn). Both annual precipitation and annual mean temperature showed high consistency between the two datasets ( figure S4).
Soil data used here include volumetric fractions of clay, sand, and silt data. This soil dataset was developed based on 8595 soil profile records compiled from the second national soil survey dataset and the 1: 1000 000 scale soil map of China (Shangguan et al 2012).
Atmospheric CO 2 concentration values were based on the annual mean CO 2 measurements at Mauna Loa Observatory, Hawaii (Keeling et al 1976). The CO 2 time series was obtained from the National Oceanic and Atmospheric Administration's Earth System Research Laboratory (http://esrl.noaa.gov/gmd/ ccgg/trends/).

Trends and detrending analyses of LAI
We assessed the trends of vegetation greenness for the period 2000-2014 using the gridded LAI dataset. For each pixel, we calculated the annual mean LAI for each year by averaging the 8 day LAI values throughout the year. We then examined the linear trend of annual LAI over the 15 year period by regressing LAI as a function of time on a per-pixel basis. The following linear regression model was used to quantify the linear trend of annual mean LAI: where y is LAI, t is time (year), and a and b are the intercept and slope, respectively. The significance level (α) of 0.05 was used for the trend analysis. Increasing LAI indicates a greening trend, while decreasing LAI indicates a browning trend.
To quantitatively assess the effects of greening and browning on annual ET and water yield, we detrended the LAI time series on a per-pixel basis, following previous studies (Wang andYou 2004, Xiao et al 2015). For each pixel, we determined the linear fit between annual mean LAI and time and then removed the linear fit from the variable: where y r is the detrended LAI, y is the original LAI, andŷ is the fitted LAI derived from equation (1). The detrended time seriesy r is essentially the residuals from the linear fit. For each pixel, we then generated the new LAI time series: where y n is the new annual mean LAI time series, and y 2000 is the annual mean LAI value in 2000. The new annual mean LAI time series ( ) y n maintains the interannual variations but has no linear trends. The residuals y r of each year were assigned to every 8 day composite based on the proportion the 8 day LAI values to the annual mean LAI values. The original and detrended 8 day LAI time series were used to drive the BEPS model, respectively.

Effects of greenness changes on ET and water yield
We conducted two model simulations on ET for China's terrestrial ecosystems over the period 2000-2014 using BEPS. The BEPS model was run at daily time step and was driven by meteorological data (temperature, precipitation, incoming solar radiation, and relative humidity), land cover type, atmospheric CO 2 concentrations, and soil properties data as described above. The original and detrended 8 day LAI time series were used to prescribe LAI for each pixel for the two simulations, and all other driving factors were exactly the same for these simulations.
For each pixel, we calculated annual ET based on the simulated daily ET values for each simulation. Water yield (mm yr −1 ) was estimated as follows: = -( ) WY P ET, 4 where WY, P, and ET are water yield, annual precipitation, and annual ET, respectively. This surface water balance approach assumes that the change in soil water and groundwater storage of catchment is negligible (Hare 1980, Sun et  The comparison of the two model simulations driven with original and detrended LAI allows us to assess how the changes in vegetation greenness affect the hydrological cycle. We compared the magnitude and spatial distribution of annual ET and water yield between the two simulations and then compared the trends of both annual ET and water yield. We assessed how the changes in vegetation greenness affected annual ET and water yield on a per-pixel basis and at the scale of large river basins (figure 1) so that we could determine how these effects varied with spatial scale.

Trends and detrending of LAI
We examined the trends of vegetation greenness for China's terrestrial ecosystems using the annual mean LAI during the period from 2000 to 2014 (figure 2). Vegetation greenness as characterized by LAI exhibited statistically significant trends over 26.1% of China's land area; pixels with significant increasing and decreasing trends in LAI accounted for 15.8% and 10.3% of China's land area, respectively. Vegetation greening mainly occurred in central and northern China. LAI significantly increased by >=0.01 m 2 m −2 yr −1 in parts of Ningxia, Gansu, Shanxi, Shannxi, Hebei, Guizhou, Chongqing, Hunan, and Hubei. Vegetation browning was mainly observed in the  Southeast, Southwest, and Northeast. LAI decreased more than 0.02 m 2 m −2 yr −1 in southeastern coastal provinces (e.g., Jiangsu, Zhejiang, and Fujian) and in southern Tibet (0.03 m 2 m −2 yr −1 ). Vegetation in Jiangxi, eastern Hunan, and parts of the Northeast also exhibited significant decreases in LAI. The nationallyaveraged LAI time series did not exhibit a significant trend during the period 2000-2014 (figure 3(a)) likely because the increases in LAI were offset by the decreases elsewhere.
We detrended LAI for each pixel, and the significant increasing or decreasing trend was removed. The detrending of the LAI time series removed the increasing trends in LAI in areas with greening trends and also removed the decreasing trends in LAI in areas with browning trends (figure 3(b)). The mean LAI value of the browning areas was about 1.0 m 2 m −2 higher than that of the greening areas. This is likely because vegetation browning mainly occurred in more productive southern regions that had higher LAI than less productive northern regions with greening trends. The nationally-averaged detrended LAI retained the interannual variations of the nationally-averaged original LAI ( figure 3(a)).

Evaluation of simulated annual ET and water yield
The ability of the BEPS model to simulate carbon and water fluxes of China's terrestrial ecosystems has been thoroughly evaluated against different sources of data (Liu et al 2015b). Here, we collected measured ET from 26 EC flux sites and runoff data from the 10 large river basins (figure 1) to evaluate the performance of BEPS in simulating ET and water yield, respectively. The 26 EC sites include 9 forest sites, 3 cropland sites, 12 grassland sites, and 2 wetland sites (Liu et al 2015b). The comparison between the simulated ET and the measured ET showed that BEPS estimated ET fairly well across sites (figure 4(a); y=0.79x+91.2, R 2 =0.77, p<0.0001).
We obtained annual runoff data for the 10 river basins over the period from 2000 to 2014 from the statistical data of 'China Water Resources Bulletin' (The Ministry of Water Resources of the People's Republic of China, http://mwr.gov.cn/zwzc/hygb/szygb/). According to the basin depiction on the hydrological yearbook of China, the mainland of China is divided into 10 major river basins: Songhua river basin (SHRB), Liaohe river basin (LRB), Haihe river basin (HAIRB), Yellow river basin (YERB), Huaihe river basin (HUAIRB), Yangtze river basin (YZRB), Southeast river basin (SERB), Pearl river basin (PRB), Southwest river basin (SWRB), and Northwest river basin (NWRB) (figure 1). The runoff data were converted to water yield with the units of mm yr −1 (referred to as measured water yield) based on the area of each basin. The simulated water yield showed excellent agreement with the measured water yield across river basins ( figure  4(b); y=0.81x+58.7, R 2 =0.81, p<0.0005). We also compared the simulated and measured water yield over the period from 2000 to 2014 for each river basin separately (figure S5). The model estimated water yield fairly well over time for most of the river basins with R 2 value greater than 0.55 for most river basins. Water yield was overestimated for basins that croplands are widely distributed such as LRB, HAIRB, YERB, and HUAIRB likely because irrigation was not considered in the model and as a result ET was underestimated for croplands in these basins. Water yield was underestimated for NWRB likely because of the errors in precipitation interpolation resulting from the complex terrain.
We also compared our simulated ET against the MODIS ET product (Mu et al 2011) and EC-MOD ET (Xiao et al 2008. The EC-MOD dataset was developed from FLUXNET observations, MODIS data streams, and micrometeorological reanalysis data using a data-driven upscaling approach (Xiao et al 2008). The magnitude and spatial patterns of simulated ET were generally consistent with those of MODIS and EC-MOD ET (figure S6). The discrepancies between simulated and MODIS ET mainly occurred in southern regions of China. Simulated ET exhibited strong linear relationships with MODIS ET (y=1.01x+96.8, R 2 =0.64, p<0.0001) and EC-MOD ET (y=0.76x+199.7, R 2 =0.58, p<0.0001) on a per-pixel basis (figure S7). MODIS ET exhibited higher values than simulated ET. Our lower ET compared to MODIS ET was supported by EC flux tower measurements. A recent validation effort showed overestimation of MODIS ET for eight EC sites across China (Liu et al 2014c).

Spatial patterns of annual ET and water yield
We used the original and detrended LAI data to prescribe LAI, respectively, in BEPS model simulations and then examined how the changes in vegetation greenness affected the magnitude and spatial patterns of annual ET and water yield in China on a per-pixel basis (figure 5). Vegetation greening generally increased annual ET (figures 5(a) and (c)). Large differences in annual ET (∼16%-20%) between the two simulations (original-detrended) were observed in regions with significant greening trends including parts of YERB. Northern HAIRB, LRB, and southern SHRB also showed large differences in annual ET (∼12%-16%). The increase in annual ET by vegetation greening was ∼5% in central YZRB and parts of HUAIRB. By contrast, vegetation browning generally decreased annual ET. Negative differences in annual ET between the two simulations (original-detrended) were sporadically distributed. Large decreases in ET (12%-16%) were observed in lower YZRB, northern SERB, and adjacent areas of upper YZRB and SWRB.
Changes in vegetation greenness also had significant effects on water yield on a per-pixel basis (figures 5(b) and (d)). Vegetation greening significantly decreased water yield, and large relative decreases in water yield (∼16%-20%) between the two simulations (original-detrended) were observed in the regions with significant increasing trends in LAI, such as YERB, northern HAIRB, LRB, and southern SHRB. By contrast, vegetation browning led to increases in water yield. Large increases of water yield (∼12%) were observed in the regions with decreasing ET. The spatial patterns of the changes in water yield between the two simulations were generally similar to those of the changes in annual ET but with the opposite direction and slightly larger magnitude ( figure 5(d)).

Trends in annual ET and water yield
We assessed the trends of annual ET and water yield during the period 2000-2014 using the two model simulations prescribed with original LAI and detrended LAI, respectively (figure 6). Annual ET generally increased in regions with greening trends and decreased in regions with browning trends ( figure 6(a)). Large positive trends in annual ET were observed in southeastern parts of LRB and eastern parts of YERB (8-12 mm yr −1 ); moderate increases in annual ET (6-8 mm yr −1 ) were sporadically distributed in southern river basins; small increases (2-4 mm yr −1 ) were observed in central YERB. Compared with the simulated ET with original LAI, the simulated ET with detrended LAI exhibited weaker increasing trends in annual ET in many regions with greening trends and decreasing trends in annual ET in some regions with greening trends ( figure 6(b)). The simulated ET with detrended LAI still increased in some regions mainly because of the increases in annual precipitation ( figure 1(b)). Annual ET exhibited decreasing trends (>8 mm yr −1 ) in regions of southern SWRB and lower YZRB with vegetation browning and smaller decreasing trends (4-6 mm yr −1 ) in other areas with decreasing LAI. The detrending of the LAI weakened the decreasing trends in annual ET in some regions with vegetation browning trends.
Modeled water yield with original LAI decreased in some areas with greening trends (figure 6(c)). In many other areas with greening trends (e.g., southern SHRB, eastern LRB, southern PRB, upper YZRB, and SERB), however, water yield increased (8-16 mm yr −1 ) (figure 6(c)) because of the large increases in annual precipitation ( figure 1(b)); the detrending of LAI led to larger increases in water yield in these regions ( figure 6(d)). Our results showed that the greening trends slowed down the increases of water yield under the changing climate. For both simulations, water yield showed large decreases in HUAIRB and southern SWRB (∼−20 mm yr −1 ), moderate decreases (∼−12 mm yr −1 ) in central parts of YZRB, and small decreases in NWRB (−4 to −8 mm yr −1 ) because of the moderate to large decreases in annual precipitation in these regions ( figure 1(b)).

Effects of greenness trends on ET and water yield at national and large river basin scales
Nationally-averaged annual ET simulated with original LAI exhibited an upward trend over the 15 year period that was not statistically significant (p=0.12) ( figure 7(a)). With detrended LAI, the simulated annual ET also showed an insignificant upward trend (p=0.16). The difference in nationally-averaged annual ET between the two simulations, however, exhibited a significant increasing trend (y=0.12x+2.01, R 2 =0.88, p<0.0001), indicating that vegetation greening increased nationallyaveraged ET. Nationally-averaged water yield exhibited no significant trend over the 15 year period for the both simulations ( figure 7(b)). The difference in nationally-averaged water yield between the two simulations, however, showed a decreasing trend (y=−0.12x−2.01, R 2 =0.88, p<0.0001). Our results indicated that the changes in vegetation greenness slightly increased annual ET and decreased water yield at the national scale. We further examined how the trends in vegetation greenness affected annual ET and water yield for the 10 large river basins (figure 8). The annual mean LAI averaged at the large river basin scale significantly increased for LRB (0.004 m 2 m −2 yr −1 , p<0.05), YERB (0.004 m 2 m −2 yr −1 , p<0.01), and HAIRB (0.006 m 2 m −2 yr −1 , p<0.01), and significantly decreased for SERB (−0.01 m 2 m −2 yr −1 , p<0.01) and SWRB (−0.004 m 2 m −2 yr −1 , p<0.001). The temporal trends of LAI were not significant for other 5 basins (HUAIRB, SHRB, YZRB, PRB, and NWRB).

Discussion
Many studies have shown that vegetation greenness has been increasing over the last three decades based on long-term NDVI and/or LAI records (Xiao and Moody 2004, Park and Sohn 2010, Li et al 2012, Chen et al 2014. Changes in vegetation greenness in China were driven by multiple environmental and human factors (Tian et al 2011, Xiao et al 2015. Air temperature was the leading climatic factor driving the increases in vegetation productivity (Zhou et al 2001, Xiao et al 2015. Increasing precipitation also enhanced vegetation productivity in semi-arid and arid regions (Xiao et al 2015). The increase in diffuse radiation (Ren et al 2013) can also enhance plant growth (Xiao et al 2015). A modeling study indicated that rising atmospheric CO 2 significantly increased China's terrestrial carbon storage (Tian et al 2011). Nitrogen deposition enhanced vegetation productivity in regions with large increases in nitrogen deposition (Tian et al 2011, Xiao et al 2015. Forest plantations were partly responsible for the increases in vegetation greenness (Lu et al 2012, Xiao et al 2015, Zhang et al 2015b. Elevated crop yield resulting from improved agricultural management practices (e.g., improved irrigation infrastructure, increasing use of chemical fertilizers, and substitution of higher-yield crops for lower-yield crops) contributed to the increases in vegetation greenness (Xiao et al 2015). The relative contributions of these environmental and human factors to increasing vegetation greenness varied with spatial scale (Xiao et al 2015).
Despite the overall greening trends, browning trends were also observed in many areas in China. Multiple mechanisms are possibly responsible for the decreasing trends in vegetation greenness. The decline in vegetation greenness was mainly caused by ecosystem degradation (Li et al 2012) and urbanization (Liu and Gong 2012). Decreasing vegetation greenness in southeastern China and parts of southwestern China was possibly caused by the climatic abnormality (droughts and ice storm) and land cover changes (i.e., urbanization) (Peng et al 2011, Liu and Gong 2012, Liu et al 2012b. Vegetation browning in northeastern China was mainly associated with negative effects of increasing droughts (Peng et al 2011, Liu et al 2013a, 2014b. The vegetation greening and browning trends observed were generally consistent with those identified in previous studies (Chen et al 2014. Our model simulations showed that on per-pixel basis, the increasing trends in LAI (vegetation greening) generally increased annual ET and decreased water yield, while decreasing trends in LAI (vegetation browning) decreased annual ET and increased water yield. The differences in annual ET and water yield between the two simulations exhibited increasing and decreasing trends, respectively, at the scales of large river basins and at the national scale, indicating that vegetation greening had positive effects on ET and negative effects on water yield at these spatial scales. Despite the effects of vegetation greening, the nationally-averaged annual ET and water yield did not exhibit significant increasing or decreasing trends. This is partly because vegetation greening trends only accounted for 15.8% of China's land area. For regions without significant changes in LAI or vegetation type, the changes in ET or water yield may follow different directions or exhibit no trends depending on the changes in precipitation and temperature. Moreover, vegetation browning was also observed in some regions, and the resulting decrease in ET and increase in water yield could partly offset the effects of vegetation greening. This indicates that the net effects of vegetation greening such as afforestation and reforestation on the hydrological cycle depend on the spatial scale. The State Forestry Administration of China expected that  . Spatially-averaged water yield for the 10 river basins during the period 2000-2014. The difference in water yield stands for the difference between the simulated water yield with original LAI and the simulated water yield with detrended LAI. the continued ecosystem restoration program will increase national forest cover to 26% by 2050 (Lei 2005). Vegetation greening therefore will likely have larger impacts on the hydrological cycle.
The effects of vegetation greening particularly afforestation projects on water yield have received growing attention. Some studies indicated that afforestation/reforestation caused significant increase of ET and reduction of streamflow , Yu et al 2009, Mátyás and Sun 2014. Forests typically have higher ET than other vegetation types, and the conversion of croplands, grasslands, and bare lands to forests can lead to higher ET. For example, a pervious study showed that vegetation restoration resulted in different reduction of average water yield in the semiarid Loess Plateau region (50%) and the tropical southern region (30%), respectively (Sun et al 2006). By contrast, another study showed that forest recovery at Guangdong Province did not cause significant water reduction . A recent global-scale synthesis study indicated that land cover changes can lead to greater hydrological responses in non-humid regions or in watersheds of low water retention capacity . Our results showed that vegetation greening including afforestation/reforestation enhanced annual ET and reduced water yield, and the effects of vegetation greening on ET and water yield could be offset by other factors at the large river basin and the national scales. Our results also showed the difference between simulated water yield with original LAI and simulated water yield with detrended LAI was larger in northern river basins than in southern river basins, which indicates that vegetation change plays a larger role in regulating the water cycle in semi-humid, semi-arid, or arid areas.
There has been a debate on the feasibility and effectiveness of afforestation/reforestation efforts in northern China (Sun et al 2006, Cao 2008, Ma et al 2013. On one hand, afforestation/reforestation has improved ecological environment and sequestered more carbon in ecosystems (Liu et al 2014a). On the other hand, afforestation/reforestation could increase regional ET and reduce water yield, thereby aggravating the water shortage issue in northern China. Some previous studies argued that the feasibility of afforestation was constrained by the water availability in China's arid and semi-arid regions (Cao 2008). The carbon gain of ecosystems is associated with the cost of water consumption. Our results showed that afforestation/reforestation increased annual ET and reduced water yield. Afforestation efforts perhaps should focus on southern China with larger water supply given the water crisis in northern China (Hao et al 2015a) and the negative effects of these efforts on water yield (Xiao et al 2013).
Afforestation/reforestation induced changes in vegetation greenness can have feedbacks to the climate by altering the hydrological and energy cycles (van Dijk and Keenan 2007, Trabucco et al 2008, Ellison et al 2012, Jiang et al 2015. A regional climate modeling study indicated that the northern China forest shelterbelt project was likely to improve overall hydroclimatic conditions by increasing precipitation, relative humidity, and soil moisture (Liu et al 2008b). Using reanalysis atmospheric perceptible water and satellite vegetation index data, Jiang and Liang (2013) found that improved vegetation greenness strengthened ET and increased summer atmospheric water vapor over Northern China, indicating that afforestation/reforestation may have positive effects on regional water resources availability. Our results showed that the effects of the changes in vegetation greenness on ET and water yield varied with spatial scale. Assessing the net effects of vegetation greening/browning on annual ET and water yield in China requires the consideration of various environmental and human factors on plant productivity, mapping of the exact location and type of forests annually, and incorporation of the feedbacks to the climate.

Conclusions
We examined how the recent trends in vegetation greenness affected annual ET and water yield of China's terrestrial ecosystems during the period from 2000 to 2014. Significant trends of vegetation greenness were identified over 26.1% of China's land area, with 15.8% for greening and 10.3% for browning areas. Vegetation greening mainly occurred in central and northern China, while browning was mainly observed in southeastern, southwestern, and northeastern China. On a per-pixel basis, greening increased annual ET and decreased water yield, while browning reduced annual ET and increased water yield. At the large river basin and national scales, vegetation greening trends had positive effects on annual ET and had negative effects on water yield. The effects of the changes in vegetation greenness on the hydrological cycle depend on the spatial extent of the region and the fraction of greening or browning area. Our results indicate that afforestation efforts perhaps should focus on southern China where precipitation is abundant given the water crisis in northern China and the negative effects of vegetation greening on water yield. Future studies on the effects of vegetation greening/ browning on the hydrological cycle at large scales should account for the feedbacks of the changes in ET and other biophysical properties (e.g., albedo) to the climate. We obtained the MODIS reflectance and land cover data from NASA (http://ladsweb.nascom.nasa.gov), MODIS ET data from the Numerical Terra dynamic Simulation Group, University of Montana, (http:// ntsg.umt.edu), EC-MOD ET from Global Ecology Group, University of New Hampshire (http:// globalecology.unh.edu), meteorological data from China Meteorological Administration (http:// cdc.cma.gov.cn), and soil data from Beijing Normal University, China (http://globalchange.bnu.edu.cn). The digital boundary of river basins in China was kindly provided by Qiu'an Zhu, Northwest A & F University. We thank the anonymous reviewers for their constructive comments on the manuscript.