1 Introduction

The atmospheric CO2 concentration has been rising and is expected to continue rising through this century at a debatable rate. Elevated CO2 concentration enhances the atmospheric greenhouse effect and can cause changes in surface temperature and distribution of precipitation. Those changes could further result in shifts in distributions of global vegetation (e.g., Emanuel et al. 1985; Smith et al. 1992). Meanwhile, elevated atmospheric CO2 stimulates the photosynthesis rate and increases carbon intake and assimilation by plants, thereby promoting plant growth (e.g., Prior et al. 2011). Increased photosynthesis rate would be accompanied by changes in plant transpiration rate. The latter can cause changes in water budget in soils and at the surface (e.g., Gerten et al. 2004). Idso and Brazel (1984) show that in an atmosphere of doubled CO2 from its current amount, vegetation in the western United States would reduce transpiration by about two thirds of its current rate. This reduction of transpiration could result in an increase of streamflow by about 40–60%. Such changes in soil and surface water availability would further feedback to and influence ecological processes, such as phenological dynamics (Band et al. 1993) and water use efficiency (Winner et al. 2004; Yu et al. 2004). It is critical to understand these changes in vegetation–hydrology interactions in order to accurately describe future water resource availability and vegetation distribution in an elevated CO2 environment (e.g., Arora 2002; Shafer et al. 2015; Sitch et al. 2008).

Responses of vegetation growth to elevated CO2 amounts differ among plant species because of their different photosynthesis pathways (e.g., Miles et al. 2004; Prior et al. 2011). Miles et al. (2004) indicate that among all 69 Angiosperm species in the Amazonia, high trees (>25 m in height) exhibit the least response to changes in CO2 amount, and species with narrow ranges and short generation times have the greatest response. Prior et al. (2011) show that plants with the C3 photosynthetic pathway often exhibit greater growth responses to CO2 change than C4 plants. Elevated atmospheric CO2 reduces plant transpiration by reducing stomatal aperture. This effect could be offset however by an increase in surface area of leaves for plants that grow faster in high CO2 environment. These changes affect water budget at the surface and in soils. As Li and Ishidaira (2012) have shown, an increase in atmospheric CO2 alone could lead to 11.9–21.8% runoff increase in humid areas (non-limited water environment) but to a huge 48.6% decrease in arid areas (water limited environment). Between humid and arid climates, ecological systems in semi-humid or semiarid climates are much more fragile, and the responses of vegetation dynamics and water balance to elevated CO2 could be quite different from that in either humid or arid climate.

One such typical semiarid environment is in the Loess Plateau in central China. Historical records indicate that the Loess Plateau endured large alternations of warm-humid and cold-dry climate at various timescales (Tan et al. 2014). In the past 2000 years, the area has suffered a steady decline of forest coverage when its climate has become more semiarid. Corresponding changes in surface vegetation type, including vegetation loss in some areas, have raised the region’s vulnerability to soil erosions and frequent extreme climate events, such as droughts and dust storms (Wang et al. 2006). It is interesting to know if this deteriorating situation might be altered by elevated atmospheric CO2 and the impacts of elevated CO2 on vegetation and surface hydrology. Such information is essential for making policies to revive or improve local environmental integrity (Xiao 2015).

The increase of the atmospheric CO2 concentration in the Loess Plateau has been at a rate of 2.2 ± 0.8 ppmv a−1 from 1991 to 2011 (e.g., Zhou et al. 2003; Fang et al. 2014). This rate is higher than the average global rate of 1.69 ppmv a−1 (MacFarling Meure et al. 2006), and could strongly affect vegetation and its interactions with hydrology in the Loess Plateau.

In this study, we quantify the effect of increasing CO2 on vegetation and surface hydrology in the Loess Plateau, using the Jinghe River basin in the Plateau as our study site. We use the Lund–Potsdam–Jena (LPJ) dynamic global vegetation model to quantify vegetation responses in different CO2 change scenarios. The effects of change in vegetation function in the elevated CO2 environment on surface water balance will be quantified. In addition, differences of those effects in semiarid and semi-humid climate conditions in the Loess Plateau will be examined to understand variations of the effects of elevated CO2 in different climates.

2 Study area and data

The Jinghe River is one of the main tributaries of the Yellow River in China. Jinghe River Basin is in the central Loess Plateau in northwestern China from 106°14′–109°06′E and 34°46′–37°24′N, covering an area of 45,373 km2. The area of the basin upstream of the hydrological station (basin drainage outlet) at Zhangjiashan is 43,216 km2 (Fig. 1a) and is the focus area of this study (hereafter JRB). The average elevation of the JRB is 1424 m above sea level.

Fig. 1
figure 1

Distributions of (a) topography, (b)mean annual precipitation, and (c) mean annual temperature in the JRB, China

Full size image

From the recent survey data of Peng et al. (2015), the JRB has 46.5% grassland, 41.6% farmland, and 10.2% forest. Forests are concentrated in semi-humid climate areas in the south and along the slopes of terrains in the southeast of the JRB (Fig. 1a and b). The northern portion of the JRB is dry and featured with loess tableland with grass and shrubs as the dominant vegetation. Over the recent history of agricultural development, suitable areas in the basin have been cultivated to grow crops, resulting in nearly 42% of crop lands in the JRB (Suo et al. 2008).

Data used in our vegetation (LPJ) model include monthly meteorological data from 1916 to 2012. These data are from the CRU TS3.23 dataset (Harris et al. 2014) and include monthly precipitation, mean temperature, rainy day frequency, and cloud cover, all at 0.5°×0.5° resolution. Because the CRU data underestimates the precipitation and overestimates the temperature of the JRB (Huang et al. 2016), we adjust the CRU data based on a local 0.5°×0.5° resolution gridded dataset, CN05, which was developed by the China Meteorological Administration. The CN05 dataset, developed from observations at more than 2472 stations in China, has an advantage in data accuracy, but covers a shorter period from 1961 to 2012. We used the 52-year CN05 data and developed their linear correlations with the CRU TS3.23 data of monthly precipitation and temperature at the same grids. Using those relationships, we adjusted the CRU monthly precipitation and temperature data from 1916 to 2012. We note that this adjustment could add uncertainties to the climate data used in this study. Effects of these potential uncertainties on our model outcomes would be expected to be small however because of very high correlations between the two datasets in their shared decades (R2 = 0.896 for precipitation and 0.996 for temperature).

Analyzing the data from 1916 to 2012, we found that the JRB averaged annual precipitation is 520.7 mm. Annual precipitation decreases from the southeast semi-humid area (annual mean of 589.5 mm) to the northwest semiarid area (annual mean of 428.9 mm) (Fig. 1b). The driest year is 1942, and the wettest year is 1964 (346.0 and 760.2 mm annual precipitation averaged in the JRB, respectively) (Fig. 2a). The mean annual temperature in the JRB decreases from the southeast semi-humid area (ranging from 6.2–12.3 °C) to the northwest semiarid area (ranging from 5.4–8.2 °C) (Fig. 1c). The annual mean temperature fluctuates between 6.8 and 9.7 °C. It is warmer before the 1950s and also after 1986 and cooler from 1950 to 1985 (Fig. 2b).

Fig. 2
figure 2

Variations in (a) annual precipitation (P), (b) annual average surface air temperature (T), and (c) annual CO2 concentration from 1916 to 2012 in JRB

Full size image

Annual atmospheric CO2 concentration data developed by the Scripps CO2 Program (MacFarling Meure et al. 2006) are used in this study. The data show that the annual mean atmospheric CO2 concentration in the study region rose from 301.6 ppmv in 1916 to 391.2 ppmv in 2012. The rise has accelerated since 1965 (Fig. 2c), especially from 1991 to 2010 when the CO2 concentration jumped from 353.2 to 387.0 ppmv. These changes are consistent with the observed rise from 355.2 to 389.5 (±1.9) ppmv measured at the international CO2 monitoring site in Waliguan (100.9°E, 36.28°N) near the JRB (Fang et al. 2014; Zhou et al. 2003).

Monthly streamflow data from 1932 to 2012 at the Zhangjiashan hydrological station are obtained from the Shanxi Hydrometric and Water Resource Bureau and used in comparison with the LPJ model output. Remote sensing products of vegetation in JRB derived from the Global Inventory Modeling and Mapping Studies (GIMMS) NDVI (Normalized Difference Vegetation Index) (1982–2012) and from the MODIS MOD15A2H-LAI (leaf area index) (2005–2012) are used to compare with modeled vegetation conditions. The soil profile data used in the model are from the Food and Agriculture Organization (FAO) soil dataset (Zobler 1986) with nine soil types.

3 Lund–Potsdam–Jena model, model validation, and experimental design

The Lund–Potsdam–Jena (LPJ) dynamic global vegetation model is a process-based approach to describe terrestrial vegetation dynamics and associated carbon and water exchanges in the terrestrial system. Details of model physics, biophysics, and dynamics are described in Sitch et al. (2003) and Gerten et al. (2004) and not repeated here.

Calibration of the LPJ model in the JRB follows the procedures described in Sitch et al. (2003). The model was integrated using data from the JRB. The data include observed climate, soils, and atmospheric CO2 concentration in the JRB averaged over the first 30 years of our study period, 1915–2012. The integration was for 1000 years to allow various carbon pools in soils and terrestrial carbon cycle that are not observed at the site to reach an equilibrium (Sitch et al. 2003). This process also yields vegetation type and composition in the JRB consistent with its climate and soils. Only at such an equilibrium could the model be used to examine responses of carbon cycle, including vegetation and hydrology dynamics, to anthropogenic and climate disturbances.

The model was further validated by comparisons of water balance between simulated and observed runoff and vegetation between simulated LAI and satellite remote sensing NDVI/LAI. In our calibration/validation, we found that the thickness of the two soil layers in the LPJ model is the most sensitive parameter influencing the model results. Our calibration suggests the same thicknesses of 1.4 m for both the soil layers in the study basin. Other model parameters suitable for the JRB are found to be similar to those suggested by Sitch et al. (2003). Details of the calibration are summarized in Huang et al. (2016).

The calibrated LPJ model simulated dominant vegetation type and distribution are shown in Fig. 3 and are consistent with that observed in the JRB (vegetation classification scheme of Prentice et al. (2011) is used in this study). In Fig. 3, temperate broad-leaved summer-green (TBS) is in the southeast of the JRB. Northward of that area, grass (C3) becomes dominant and is mixed with shrubs and patches of short woody plants/trees (mostly boreal needle-leaved evergreen, BNE), before becoming grass only in the northern tip of the JRB. This pattern largely resembles the actual land-cover (contoured areas numbered 1–4 in Fig. 3) that has TBS in the southeast JRB, more grass mixed with shrubs in the main body of the basin, and grass only in its northern tier. The major differences in model simulated and actual land-cover are along the east fringes of the JRB, where the dominant BNE and BBS (boreal broad-leaved summer-green) plant types along the slopes of terrains are not simulated as the dominant plant types. Because in those areas the model also has BNE and BBS in the vegetation mix but as lesser dominant types than grass, these differences in model simulated vegetation in the JRB are considered small and acceptable.

Fig. 3
figure 3

Color code shows PFTs simulated by the LPJ model. Contour lines show the boundaries of the observed land cover types with the region 4 for grass mixed with crops

Full size image

The calibrated LPJ model with the land-cover is used to simulate JRB runoff from 1965 to 2012. (Our production integration is from 1965 to 2012 because the atmospheric CO2 concentration before 1965 remains similar to the value used in the calibration, 1915–1945.) Comparisons of runoff between the simulation and observation at the Zhangjiashan hydrological station (outlet of the JRB) show that the coefficient of determination (R2) is 0.36 for annual runoff (Fig. 4a) and R2 = 0.7 for average monthly runoff (Fig. 4b); both significant at the 99% confidence level. While the statistics of the simulated runoff are strong, there are some large deficiencies between the simulated and actual runoff. For example, the simulated annual runoff loses strong interannual fluctuations shown in the observed runoff in some periods (Fig. 4a). The average annual hydrograph from the simulation has more runoff in spring months and also a near one-month delay in peak runoff (Fig. 4b). These differences could affect model results related to those particular aspects and, because of such, they should be interpreted with caution.

Fig. 4
figure 4

Observed and (a) simulated annual runoff and (b) 1965–2012 averaged monthly runoff of JRB. Annual precipitation is shown in (b) by the scale on the right axis

Full size image

While comparing annual variations of simulated LAI with observed NDVI from 1982 to 2012 and available LAI from 2005 to 2012 (Fig. 5), we found that they match well during 1982–1994 and 2005–2012. Relatively large discrepancies exist from 1995 to 2004 primarily because of changes of cultivated areas in the JRB resulting from regional economic policy changes. In the mid-1990s, farmers were given the freedom to migrate to cities to find jobs, and many of them did. That migration affected land-cover in the following years. Part of those farmers returned to their farms to receive subsidies when a “Grain for Green” policy was initiated in 1999. In subsequent years, that policy resulted in an increase of woody vegetation in some previous farm lands (Geng et al. 2008).

Fig. 5
figure 5

Simulation of leaf area index (LAI) and observation from GIMMS-NDVI and MODIS-LAI satellite production

Full size image

After these calibration and validation, we apply the LPJ model to the JRB to study basin-scale vegetation and water balance responses to the rising CO2 amount in the atmosphere. Two model experiments are carried out. Both are integrated from 1965 to 2012 because most of the CO2 increase took place after 1965 (Fig. 2c) when the climate data are also most reliable. One experiment uses the LPJ model to simulate vegetation dynamics and interactions with water balance at a fixed-lower CO2 concentration in the atmosphere. The fixed-lower CO2 amount is 318.9 ppmv, observed in 1964. The other uses the observed rate of increase in atmospheric CO2 concentration from 1965 to 2012. Because climate conditions in these experiments are identical, their differences in vegetation condition and water balance in JRB will help distinguish effects of the rising concentration of atmospheric CO2.

4 Effects of elevated CO2 on vegetation dynamics and water balance

In evaluating the effects of elevated atmospheric CO2 on vegetation and surface water balance, we use model simulated plant characteristics, e.g., LAI, foliage projected cover (FPC), and net primary production (NPP, gross primary production less respiration cost). Among model outputs of hydrological variables used in our analyses are monthly and annual runoff and actual evapotranspiration (E), which is the sum of plant transpiration (ET), bare soil evaporation (ES), and evaporation of plant intercepted water (EI). The average of any of these variables over the JRB is calculated using the grid areal weighted averaging method.

4.1 Basin-averaged effects of elevated CO2 in 1965–2012

Model results summarized in Table 1 show that following the rising atmospheric CO2 from 1965 to 2012 trees are becoming more dominant than grass in the land-cover of the JRB. This change is evident in that LAI, FPC, and NPP increase for TBS, BNE, and BBS, but decrease for C3 (grass). Compared to the results of model simulation using fixed-lower CO2, the basin average annual LAI, FPC, and NPP increase by 8.4, 0.7, and 6.0%, respectively (Table 1). The time series of NPP and annual E and runoff (R) are shown in Fig. 6b–d for the rising CO2 and fixed-lower CO2 simulations. The difference of NPP between the two simulations enlarges following the rise of atmospheric CO2 (Fig. 6b). For example, from 2000 to 2012, the mean annual NPP of the JRB increases by 10.6% from NPP in the fixed-lower CO2 case (Table 2).

Table 1 Mean annual vegetation and water balance indicators of the JRB from the two simulations (1965–2012). Average values for the semiarid and semi-humid area of the JRB are also included. [relative difference = (rising CO2 – fixed-lower CO2)/ fixed-lower CO2]
Full size table
Fig. 6
figure 6

Three-year moving average of (a) observed precipitation (P), (b) simulated NPP, (c) actual evapotranspiration (E), and (d) runoff (R) for the rising (dash-line) and fixed-lower CO2 (gray-line) cases. Blue and black histograms show, respectively, difference and relative difference between the rising CO2 and fixed-lower CO2 results. Theshaded area indicates the two periods (1977–1984 and 1985–1990) used in the detailed analysis

Full size image
Table 2 The same as Table 1 but for results averaged over 2000–2012
Full size table

Figure 6b also shows that while NPP increases following the rise of atmospheric CO2, the fluctuation of NPP follows the variation of local climate, especially precipitation (cf. Fig. 6b and a). This result indicates that although a richer CO2 environment encourages plant growth by increasing photosynthetic uptake, the actual growth in individual years is still dependent on water availability (precipitation). The climate limitation on NPP is caused by stomata closure of plants in response to drier climate. While this process reduces plant water loss, it slows plant photosynthetic uptake of CO2. As shown in Drake et al. (2017), however, this latter effect can be offset to some extent by the increase of internal CO2 partial pressure. Such effect is also evidenced in our result (upper histogram in Fig. 6b) by the larger relative difference of NPP in dry climate.

Compared to the strong response of NPP, the response of surface water balance to rising atmospheric CO2 is small (cf. Fig. 6b–d). This weak sensitivity of water balance to rising CO2 results from some cancellations among different processes contributing to E in the elevated CO2 environment. Specifically, a decrease of ES and an increase of EI (Fig. 7b and c) contribute to an increase in vegetation growth and thus higher LAI and FPC. Plant transpiration ET decreases in most years but increases in some very dry years. These are suggested by the positive differences of ET in Fig. 7a when compared to the fixed-lower CO2 case. The change of ET further amplifies with rising atmospheric CO2. In the LPJ model, ET is determined from ET = Min[S, D] × fv. In this formula, S is plant- and soil-limited water supply function, and D is an atmosphere-controlled demand function that is a strong function of potential canopy conductance (Federer 1982). The parameter fv is the fraction of vegetation cover in a grid cell. In the well-watered condition (D < S), ET decreases in the elevated CO2 environment because of a decrease in canopy conductance (gc) (Gerten et al. 2004). In water-limit condition (D≥S), ET increases in the elevated CO2 environment primarily because of an increase in vegetation coverage (fv) at higher rates of photosynthesis (Keenan et al. 2013).

Fig. 7
figure 7

Simulated annual values of (a) ET, (b) ES, (c) EI, and (d) WUE for the rising CO2 (dash-line) and fixed-lower CO2 (gray-line). Blue and black histograms show, respectively, the difference and relative difference between the rising CO2 and fixed-lower CO2 results

Full size image

Our simulated results show that from 1965 to 2012, the averaged annual ET in the JRB deceased by 2.3 mm (−0.7%) in the rising atmospheric CO2 simulation compared to the fixed-lower CO2 run. This decrease of ET is a net result of a decrease in grass ET (−6.9 mm) and an increase in ET from woody vegetation (trees) (4.6 mm), when grass (C3) shifts to trees (TBS, BNE, and BBS) in the elevated CO2 environment. The decreased ET from grass is contributed by decreases of 6.3 and 0.6 mm (91.3 and 8.7% of the decreased ET) due to decreases in vegetation coverage (fv) and canopy conductance (gc), respectively. The increased ET from trees is contributed by increases of 3.2 and 1.4 mm (69.6 and 30.4% of the increased ET) due to increases in vegetation coverage (fv) and canopy conductance (gc), respectively. These changes indicate negative and positive gc responses to elevated atmospheric CO2 for grass and trees in the study region. Increases of gc in an elevated CO2 environment have also been reported in hot and dry biomes in dry environments (Purcell et al. 2018).

Meanwhile, ES decreases from 1965 to 2012 by 4.3%, and EI increases by 6.7% in the JRB (Table 1). Together, these changes result in a slight decrease of E in the rising atmospheric CO2 simulation compared to the fixed-lower CO2 run. Consistent with this slightly reduced E, mean annual runoff R increases slightly with the rise of atmospheric CO2. The average increase of R is 1.7% relative to the fixed-lower CO2 case (Table 1). We also note large fluctuations of R especially in dry years. In those years, increased water consumption by vegetation growth in the elevated CO2 environment causes R to decrease. As an example, in the dry year of 2007 simulated R is 5.8% lower in the elevated CO2 case than the fixed-lower CO2 case (Fig. 6d).

4.2 Effects of elevated CO2 in semiarid and semi-humid areas in the JRB

We further examine variations of the response of vegetation and surface water balance to rising atmospheric CO2 across different climate zones in the JRB. On the basis of the strong north-south precipitation gradient (Fig. 1b), we divide the JRB into a semiarid region north of the 500 mm annual precipitation contour line (the boundary between the blue and light blue zones in Fig. 1b) and a semi-humid region south of that line. In addition, we compare the responses of water use efficiency WUE (WUE = NPP/ET) in these two climate regions. It is noted that because the climate variables driving the LPJ model are the same in the rising CO2 and fixed-lower CO2 simulations, no indirect effect of CO2 rise on these responses through its effect on the climate is measured.

Table 1 summarizes the responses of LAI, NPP, E, and R to the rising atmospheric CO2 from 1965 to 2012 in the semiarid region of the JRB. Compared to the results from the fixed-lower CO2 simulation, NPP and LAI in the elevated CO2 run show an increase of 7.6 and 11.1%, respectively, in the semiarid region. This increase is limited to tree species, i.e., TBS, BNE, and BBS, however. The largest increase is seen in BNE (more drought resistance species) and the smallest in TBS. NPP and LAI of C3 (grass) decrease in the elevated CO2 case. Because the decrease in C3 is small, the averaged NPP and LAI increase in the semiarid region.

An intriguing difference in the semiarid region is between the large increase in LAI and NPP and rather small changes in E in response to the rise of atmospheric CO2 (Table 1). In fact, the mean of annual E changes little between the rising CO2 and fixed-lower CO2 cases. The small changes in E is a net result of quite different responses of the components constituting E, i.e., EI, ET, and ES, in their responses to the CO2 increase. In Table 1, EI shows a considerable increase in the rising CO2 case. This increase could result from expansion of woody (tree) vegetation (BNE, BBS, and TBS) in the elevated CO2 environment. Consistently, evaporation from bare surfaces, ES, is 11.4% lower in the rising CO2 than in the fixed-lower CO2 case. These changes nearly offset one another, thus yielding a rather small net positive change in E, which also explains a slight reduction of R in the semiarid region of the JRB following the rise of CO2.

Changes of these vegetation and water budget components and their net effects in the semi-humid region of the JRB are also summarized in Table 1. Results in Table 1 show a 5.2% increase of NPP in the rising CO2 case from the NPP of the fixed-lower CO2 case. This amount of increase is smaller than the increase of 7.6% in the semiarid region (Table 1). The increase of NPP in the semi-humid region in the rising CO2 case is also attributed to an expansion of tall woody vegetation and small contraction of C3 plants. Furthermore, from analyzing the NPP budget, we find that the responses of BNE and BBS to rising CO2 is mild. A large change is found in TBS species (Table 1). The net increase of NPP in the semi-humid region is smaller than in the semiarid region with the same rising CO2 rate because, according to Miles et al. (2004) and Tricker et al. (2009), the short rotation species, BNE as well as C3, in the semiarid region are more sensitive to the rising atmospheric CO2 than trees of TBS and BBS in the semi-humid region (also see the distribution of plant functional types in Fig. 3).

Total evapotranspiration E in the semi-humid region is slightly smaller in the rising CO2 case than in the fixed-lower CO2 case because of reduced ET and ES. The reduction of ET is particularly large. The reduced E explains the slight increase of R in the semi-humid region of the JRB (Table 1).

The differences of surface water balance and vegetation growth between the semiarid and semi-humid regions are amplified in a higher CO2 environment. As shown in Table 2, for the average value during 2000–2012 when the atmospheric CO2 concentration rose to the highest in the study period, the increase of LAI and NPP in the rising CO2 case are larger in the semiarid area (22.4 and 13.1%, respectively, relative to the fixed-lower CO2 case) than in the semi-humid area (13.9 and 9.3%, respectively). The stimulated plant growth in the high CO2 environment consumes more water (0.6% increase of E and 11% decrease of R) in the semiarid region. Less water is used in the semi-humid area (0.5% decrease of E and 4.9% increase of R).

The changes of NPP and ET caused by rising CO2 define the change of water use efficiency, WUE(=NPP/ET). Our further analysis reveals a constraint of CO2 on those changes such that WUE remains nearly invariant in the semi-humid and semiarid climate regions under the same CO2 level. In the rising atmospheric CO2 case, the average WUE over 1965–2012 is 1.43 gC/kg H2O in both sub-climate regions. The same WUE of 1.34 gC/kg H2O is also obtained in the two different sub-climate regions in the fixed-lower CO2 case (Table 3). Additional evaluations of the NPP and ET data averaged in the recent higher CO2 concentration period from 2000 to 2012 show a higher but still constant WUE = 1.48 gC/kg H2O across the different climate regions in the JRB. It may be particularly intriguing that WUE remains near the constant of 1.34 gC/kg H2O in those years for the fixed-lower CO2, as in the prior years in the fixed CO2 run, even when the climate input has changed considerably. These results, summarized in Table 3, suggest that plant WUE would increase primarily following the rise of the atmospheric CO2 concentration, while climate effects are on interannual fluctuations of plant growth and ET. Those changes are, as suggested by our results, kept near a constant WUE specified by the atmospheric CO2 concentration: less NPP in drier years with proportionally reduced ET and more NPP and ET in wetter years (Figs. 6a, b, and 7a). Higher WUE in elevated atmospheric CO2 is attributed to smaller ET in the semi-humid region, and it is attributed to larger ET in the semiarid climate (ET is smaller/larger in the elevated CO2 than in the fixed-lower CO2 for the semi-humid/semiarid climate area, as shown in Table 3). Higher WUE with increase in ET has also been found at three FLUXNET sites (Keenan et al. 2013).

Table 3 Mean annual NPP, ET, and WUE for the period of 1965–2012 and 2000–2012
Full size table

5 Discussions and concluding remarks

Rising atmospheric CO2 concentration stimulates plant photosynthesis while often, but not always, reducing plant stomatal aperture and conductance (e.g., Saxe et al. 1998; Farquhar 1977). The subsequent increase in carbon uptake and assimilation by a plant enhances its growth and water use efficiency. These processes would affect the growth of plants and can further cause changes in vegetation composition and consequently the surface water balance. In this study, we examined these changes in the Jinghe River Basin (JRB) in the Loess Plateau of central China, using the Lund–Potsdam–Jena (LPJ) dynamic global vegetation model. After calibrating and validating the model to the JRB, we analyzed our model results from two simulations, both from 1965 to 2012: one using the observed rise in atmospheric CO2 amount and the other using a fixed lower CO2 concentration observed in 1964.

Results from analyses of the model simulated data show a significant increase in vegetation growth in the JRB from 1965 to 2012 following the rising atmospheric CO2 concentration. The average NPP and LAI are 10.6 and 16.3%, respectively, higher in the rising atmospheric CO2 simulation than the fixed-lower CO2 run, averaged over 2000–2012 when the CO2 level is the highest in recent decades (average 379 ppmv). While these results reiterate the enhanced fertilization effect of elevated atmospheric CO2 on vegetation growth (e.g., Prior et al. 2011; Swann et al. 2016), additional effects of rising atmosphere CO2 are found to change the vegetation composition in the JRB. Our results indicate an increase of woody (tree) vegetation (more dominant among grid cell vegetation species) and a decrease of C3 (grass) following the rise of atmospheric CO2.

The basin averaged change of water budget between the two simulations shows a slight decrease in total evapotranspiration, E, and an increase in runoff, R, in the elevated CO2 run in the study period (1965–2012). Further examinations of individual terms in the budget of E (which is a major source/sink for R) reveal that plant transpiration ET generally decreases following the rise of atmospheric CO2, a result suggesting an increase in plant water use efficiency (WUE) in an elevated CO2 environment. This model result is consistent with prior findings (e.g., Gerten et al. 2004) and is attributed to shifts of grass (C3) to trees (TBS, BNE, and BBS) in an elevated CO2 environment.

Our further examinations of vegetation dynamics and surface water budget in semiarid versus semi-humid climate areas in the JRB indicate some opposite CO2 effects. We found that in elevated atmospheric CO2 condition plants can sustain growth in a semiarid (water-limited) climate. This is because, as shown in Drake et al. (2017), a higher level of intracellular CO2 may mitigate the effects of droughts and reduce the effect of aridity on some plants through increased WUE. As a result, plants would consume more water in a drier climate of elevated atmospheric CO2 and enhance the decrease of surface runoff. The mean annual runoff decreases by 11.0% (relative to the runoff in the fixed-lower CO2 case) in the semiarid area of the northern JRB in the high CO2 concentration condition from 2000 to 2012.

On the other hand, in the semi-humid region of the southern JRB, the NPP also increases in the elevated CO2 case but at a rate smaller than in the semiarid north. ET and total surface evapotranspiration E decrease slightly compared to the fixed-lower CO2 case. This decrease leads to a small increase of the runoff in the semi-humid climate, in contrast to the decrease of runoff in the semiarid climate in the elevated CO2 environment. This asymmetry of the effects of elevated atmospheric CO2 could have exacerbated surface drying in the semiarid climate, while enhancing surface wetness in the semi-humid climate.

An increase of WUE (=NPP/ET) has been a known result of rising atmospheric CO2 (e.g., Keenan et al. 2013), although its cause remains in debate. While early experiments emphasized the effects of either an increase in NPP or a decrease in ET on WUE of various plant species (e.g., Gunderson et al. 1993; Rogers et al. 1994), recent studies have shared the consensus that the increase of plant WUE in an elevated CO2 environment results from CO2 effects on both NPP and ET (e.g., Keenan et al. 2013). Our analysis of the asymmetry of the rising CO2 effect on amplifying extreme hydrological conditions in dry and wet climate indicates that an increase in WUE primarily follows the rise of atmospheric CO2, and it is not sensitive to wet or dry climate in our study region. This result suggests a biophysical constraint of the atmospheric CO2 concentration on plant growth and Earth’s vegetation environment and offers a plausible explanation of increasing extreme conditions in the climate of fast rising atmospheric CO2.

It is recognized that our result of a nearly invariant WUE in semiarid and semi-humid climates under a given atmospheric CO2 concentration and underlined biophysical and phenological processes are derived from this one study basin and based on a single (LPJ) model. Limitations of the LPJ model, particularly in its absence of groundwater and topographic effects (e.g., loess tables and gullies in the JRB) on re-distribution of soil moisture and lateral flows, pose uncertainties on the validity of this constraint and applicability of certain results from this study. These results need to be further validated in regions of different latitude, longitude, and climate, and by more advanced dynamic vegetation models before being proven as biophysical properties of the Earth system.