Introduction
Arid and semiarid biomes cover approximately 40 % of the Earth's
terrestrial surface (Fernández, 2002). Previous studies have shown that
more than 50 % of precipitation (P) is consumed by
evapotranspiration (ET) (Yang et al., 2007; Liu et al.,
2002). Moreover, a slight change in ET could have
significant influences on water cycle and the ratio of ET/P
could increase to even 90 % or more in these regions (Mo et al., 2004;
Glenn et al., 2007). In terms of physical processes, ET
is affected by net radiation (Valipour et al., 2015), water vapor pressure
deficit (Zhang et al., 2014), wind speed (Falamarzi et al., 2014), and soil
water stress (Allen et al., 1998). Moreover, vegetation condition is also a
crucial factor influencing ET (Tian et al., 2015; Wang
et al., 2011; Piao et al., 2006; Mackay et al., 2007).
Vegetation change mainly includes phenological change (temporal) and land
use/cover change (spatial). Phenological change reflects the response of
plants to climate change (vegetation greening and browning processes) (Ge et
al., 2015), which actively controls ET through internal
physiologies such as stomatal conductance (Pearcy et al., 1989), as well as
the number and sizes of stomata (Turrell, 1947). In general, transpiration is
directly proportional to stomatal conductance at the leaf scale (Leuning et
al., 1995). At the canopy scale, ET is positively
proportional to surface conductance, which is an integration of stomatal
conductance and leaf area (Ding et al., 2014). Thus, as a good indicator of
vegetation phenological change, many studies have found that
ET is positively related to vegetation indexes such as
the Normalized Difference Vegetation Index (NDVI) (Gu et al., 2007). Land
use/cover change influences ET by modifying vegetation
species with different transpiration rates, radiation transfers within the
canopy (Martens et al., 2000; Panferov et al., 2001), topography (Lv et al.,
2006), albedos (Zeng and Yoon, 2009), soil texture (Maayar and Chen, 2006),
litter coverage (Wang, 1992), and biological soil crusts (BSCs) (Yang et al.,
2015; Fu et al., 2010; Liu, 2012). These complex processes result in no
consensus on the effects of land use/cover change on ET.
For example, during the land degradation process, some researchers found that
warming air temperature was the main cause of making ET
increase (Zeng and Yang, 2008; Li et al., 2013; Feddema and Freire, 2001). By
contrast, a decline in ET was found along with the
deforestation process because of less transpiration (Snyman, 2001; Souza and
Oyama, 2011) or higher albedo (Zeng et al., 2002). Moreover, no changes in
ET during the land degradation process were reported
either (Hoshino et al., 2009). Thus, there has been an important push to
better understand how ET responds to vegetation change,
especially to the land use/cover change.
Three methods were usually employed to assess the effects of vegetation
change on ET: numerical models, paired comparative
approaches, and in situ field observations. In these methods, numerical
models are widely used (Twine et al., 2004; Kim et al., 2005; Li et al.,
2009; Cornelissen et al., 2013; Mo et al., 2004). However, model
parameterization of vegetation conditions is a big challenge, as the
aforementioned complex underlying mechanisms may not be completely considered
in the models. Therefore, the simulated effects of vegetation change on
ET are highly dependent on model parameterizations,
which may induce uncertainty (Cornelissen et al., 2013; Li et al., 2009). The
paired comparative approach is often considered the best method; nonetheless,
it is difficult to find two sites with similar meteorological conditions but
different vegetation conditions (Li et al., 2009; Lorup et al., 1998).
Moreover, the method of in situ field observations is widely used to
investigate long-term land–atmosphere exchanges. However, the land use/cover
conditions at sites are generally stable, and only the response of
ET to vegetation phenological change can be observed,
such as the ET variations in grassland (Y. Zhang et al.,
2005), mixed plantation (cork oak, black locust, and arborvitae) (Tong et
al., 2017), vineyard (Li et al., 2015), and grazed steppe (Chen et al., 2009;
Vetter et al., 2012). Continuous field observations under both land
degradation and vegetation rehabilitation processes have rarely been
documented, especially in the semiarid shrubland.
The Mu Us Sandy Land is a semiarid shrubland ecosystem on the northern margin
of the Loess Plateau in China. The area covers only 40 000 km2 (Dong
and Zhang, 2001) and is ecologically fragile (Yang et al., 2007). In such an
ecosystem, sand dunes and BSCs are commonly observed (Gao et al., 2014; Yang
et al., 2015; Li and Li, 2000; Liu, 2012). Due to the existence of BSCs and
dry sand layers (Z. Wang et al., 2006; Feng, 1994; Liu et al., 2006; Yuan et
al., 2008), soil evaporation has been effectively retained; therefore, the Mu
Us Sandy Land contains abundant groundwater (Li and Li, 2000). During the
past decades, rapid land use/cover changes have occurred in this region due
to agricultural reclamation (Wu and Ci, 2002; Ostwald and Chen, 2006; Zhang
et al., 2006), leading to dramatic changes in vegetation conditions. With
respect to the specific question of whether land use/cover change will lead
to increases in ET or not, a continuous measurement of
ET under different land use/cover conditions is required
in this region. Coincidentally, two processes of land use/cover changes (land
degradation and vegetation rehabilitation) have occurred at the edge of the
Mu Us Sandy Land, providing us with a unique opportunity to study the effects
of land use/cover change on ET.
Hence, based on the 4-year measurement of ET by eddy
covariance techniques, this study analyzed the seasonal and inter-annual
variations in ET, and discussed the possible reasons for
the responses of ET to land use/cover change. Our
results were expected to provide a scientific reference for the sustainable
management of regional land and water resources in the context of intensive
agricultural reclamation.
Case study and data
Site description
The study was conducted at the Yulin flux site (38∘26′ N,
109∘28′ E, 1233 m),
which was established in June 2011. This site is located in a landform
transition zone that changes from the Mu Us Sandy Land to the north Shaanxi
Loess Plateau (Fig. 1). This site is a semiarid area with temperate
continental monsoon climate. According to long-term climate data (1951–2012)
from a meteorological station in Yulin (Fig. 1), the annual precipitation
varied from 235 to 685 mm, with a mean of 402 mm, and more than 50 % of
annual precipitation fell in the monsoon season (July–September). The mean
annual air temperature was 8.4 ∘C over the past 61 years. The
dominant soil type is sand (98 % sand) (saturated soil water content of
0.43 m3 m-3, field capacity of 0.16 m3 m-3, residual
moisture content of 0.045 m3 m-3). There are widely distributed
fixed sand dunes and semi-fixed sand dunes around the site, and the depth of
the dry sand layer is 10 cm (Z. Wang et al., 2006). The mean groundwater
depth at our study site from 1 July 2011 to 30 June 2015 was 3.5 m.
Location of the Loess Plateau and map of the study site (LP: the
Loess Plateau; black triangle: flux tower; white triangle: Yulin
meteorological station; (1): Tu River; (2): Yuxi River; (3): Yellow River).
![]()
Shortage of water is the critical limiting factor for vegetation growth in
this site, and drought-enduring vegetation (e.g., shrubs) prevails as a
result of droughts (Wang et al., 2002; Wu, 2006). The study site is mainly
covered with mixed vegetation: the native drought-enduring shrubs with low
water demand (e.g., Artemisia ordosica and Salix psammophila) (Fig. 2a) and the sparse grass (mainly distributed at the
bottom of sand dunes because of the better soil moisture condition) (Lv et
al., 2006). The maximum root depth of the shrubs was approximately 160 cm.
Xiao et al. (2005) reported that the growing season of Artemisia ordosica and Salix psammophila spanned from late April to late
September. Therefore, we defined the period from 1 May to 30 September as the
vegetation growing season for data analysis in this study. On 15 August and
7 September 2011, we did surveys of the vegetation coverage by randomly
selecting seven samples around the flux tower
(5 × 500 cm × 500 cm and
2 × 1000 cm × 1000 cm). We found that the vegetation
coverage was 28.2 % in August and 27.9 % in September.
Land use/cover conditions at the study site: (a) the natural land
use/cover condition of shrubland (photo was taken on 6 August 2011); (b) the natural land use/cover condition of grassland (photo was taken on
7 September 2011); (c) the undisturbed zone (natural vegetation) and the
disturbed zone (bare soil) in the land degradation process (photo was taken
on 26 April 2013); (d) the undisturbed zone (natural vegetation) and the
disturbed zone (grassland) during the vegetation rehabilitation process
(photo was taken on 16 August 2014).
![]()
At the end of June 2012, the land use/cover condition around the eastern
portion of the flux tower began to be changed by farmers (leaves and branches
were cut, and the sand dunes were bulldozed) (Fig. 2c), converting part of
the natural vegetated land to bare land, with the planning of planting
potatoes in the future. As time went on, natural grass gradually grew out in
the area of bare land before potatoes were planted. Thus, our study period
(1 July 2011 to 30 June 2015) was divided into four periods according to the
land use/cover conditions: (a) Period I (1 July 2011 to 30 June 2012), the
period with the natural land use/cover condition (i.e., mixed sparsely
distributed shrubs and grass) (Fig. 2a and b); (b) Period II (1 July 2012 to
30 June 2013), the transitional period when the land use/cover condition
started to change (some natural vegetation removed and sand dunes bulldozed);
(c) Period III (1 July 2013 to 30 June 2014), the period when the land
use/cover condition constituted two parts: the natural vegetation zone and
the bare soil zone (Fig. 2c); and (d) Period IV (1 July 2014 to
30 June 2015), the period when the bare soil zone was gradually covered by
regrowing grass (Fig. 2d).
Field measurements
Eddy covariance system measurements
Net exchange of water vapor between atmosphere and canopy at this site is
measured by the eddy covariance (EC) flux measurement, which assesses the
fluxes of land–atmosphere (such as water and energy) (Baldocchi et al.,
2001). The data are essential for the estimation of the water and energy
balance (Franssen et al., 2010). At our site, the EC system is installed at a
height of 7.53 m above the ground surface, using CSAT3 three-dimensional
sonic anemometers (Campbell Scientific Inc., Logan, UT, USA) for wind and
temperature fluctuation measurements and a LI-7500A open-path infrared gas
analyzer (LI-COR, Inc., Lincoln, NE, USA) for water vapor content
measurement.
Other measurements
Net radiation (Rn) is measured by a net radiometer
(CNR-4; KIPP&ZONEN, Delft, the Netherlands), including four radiometers
measuring the incoming and reflected short-wave radiation (RS),
and incoming and outgoing long-wave radiation (RL). Sunshine
duration (DS) is measured by a sunshine recorder (CSD3;
KIPP&ZONEN, Delft, the Netherlands). Wind speed and direction (05103,
Young Co. Traverse City, MI, USA) are measured at 10 m above the ground
surface. Precipitation (P, mm) is recorded with a tipping bucket
rain gauge (TE525MM; Campbell Scientific Inc., Logan, UT, USA) installed at a
height of 0.7 m above the ground surface. Air temperature (Ta)
and relative humidity (RH) are measured by a temperature and
relative humidity probe (HMP45C; Campbell Scientific Inc., Logan, UT, USA) at
a height of 2.6 m above the ground surface. Soil water content (θ)
is measured by time domain reflectometry (TDR) sensors (CS616; Campbell
Scientific Inc., Logan, UT, USA), soil temperature (Ts) is
measured by thermocouples (109; Campbell Scientific Inc., Logan, UT, USA),
and soil heat flux (G) is measured by heat flux plates (HFP01SC;
Campbell Scientific Inc., Logan, UT, USA) at a depth of 0.03 m below the
ground surface. These ground variables (G, θ, Ts) are
measured beneath the surface at two profiles: a plant canopy profile and a
bare soil profile. θ and Ts are measured at depths of
5, 10, 20, 40, 60, 80, 120, and 160 cm below the ground surface. The
groundwater table is measured by an automatic sensor (CS450-L; Campbell
Scientific Inc., Logan, UT, USA), which is installed in a groundwater well
close to the tower.
Flux data processing
The 10 Hz three-dimensional wind speed and water vapor concentrations that
were collected by the EC technique were processed to half-hourly latent heat
flux (λET) using Eddypro processing software (v5.2.0,
LI-COR, Lincoln, NE, USA). The main principle is that λET can be expressed as ρaw′q′‾ (where
w′ is the fluctuation of vertical wind speed, q′ is the fluctuation of
specific humidity, and
ρa is the air density). The software also applies the quality
control of data, including spike removal, tilt correction, time lag
compensation, turbulent fluctuation blocking, and spectral corrections. The
percentages of half-hourly λET values removed (including
missing and rejected) through the quality control procedure were 17.3 %
in Period I, 20.2 % in Period II, 16.5 % in Period III, and
18.6 % in Period IV. Almost all the removed λET
values occurred during the nighttime (89.1 % in Period I, 91.3 % in
Period II, 92.6 % in Period III, and 88.7 % in Period IV). During the
nighttime, the change in λET was small, and
ET values were close to zero. Therefore, after removal of the
nighttime λET values, the errors of the gap-filled
nighttime values based on the neighboring good data were small. Moreover,
nighttime λET values accounted for only a small
proportion of the daily ET. Furthermore, the percentages of
rejected and missing data in our study are similar to those reported by other
scholars, and these percentages are in a range of 15–31 % (Falge et al.,
2001; Wever et al., 2002; Mauder et al., 2006). Therefore, the λET data set was considered reliable after a quality control
procedure.
After quality control, missing and rejected data were gap-filled in order to
create continuous data sets. Three methods were applied in the gap-filling
procedure: (1) linear interpolation was used to fill gaps of less than 1 h
by calculating an average of the values before and after the data gap;
(2) for gaps that are larger than 1 h but smaller than 7 days, the mean
diurnal variation (MDV) method (Falge et al., 2001) was used; (3) for gaps
that are larger than 7 days but smaller than 15 days in daily λET values, we fitted the relationship between daily λET (W m-2) and the daily available energy flux
(Rn-G) (W m-2) in each period. We chose the function
f with the highest coefficient of correlation (R) in each period
(Yan et al., 2013), and the function was expressed as f=a⋅Rn-G2+b⋅Rn-G+c (Period I: a= 0.0014 m2 W-1, b= 0.075, c= 10.69 W m-2, R= 0.77; Period II: a= 0.0012 m2 W-1, b= 0.056, c= 17.69 W m-2, R= 0.67; Period III: a= 0.0014 m2 W-1, b= 0.16, c= 13.24 W m-2, R= 0.75; and Period IV: a= 0.0015 m2 W-1, b=-0.083, c= 25.87 W m-2, R= 0.69). Then, we used the fitted function f in each period to estimate
the daily λET values of large gaps. In addition, gaps
that are larger than 7 days but smaller than 15 days mostly appeared in the
winter, which accounted for a small proportion of annual λET.
Results
Footprint and energy balance closure
Based on the footprint model, we got the half-hourly scatter data (Eq. 2),
and, according to the wind rose diagram (Fig. 3a), the prevailing wind
directions at this site were northwesterly and southeasterly. Therefore, we
chose an ellipse to enclose the scatters and simulate the footprint
(Fig. 3b). Under unstable conditions, 93 % of the half-hourly flux data
are plotted within the ellipse.
Additionally, we measured the boundary of zone B in October 2013 when the
land use/cover condition in zone B had stopped changing (Fig. 3b). There
were 11 pixels (250 m × 250 m) in zone A and 19 pixels (250 m × 250 m) in zone B,
and thus, when calculating the weight-averaged
NDVI (NDVIw) within the footprint, we chose the
weighted coefficient as β=11/(11+19).
Diagrams of the wind rose and footprint: (a) wind rose of
the study site by using half-hourly wind speed and wind direction data
and (b) simulated footprint by ellipse (the long axis is 1682 m,
and the short axis is 1263 m; zone A is the source area in which the land
use/cover condition did not change, while zone B is the source area in which
the land use/cover condition did change due to human activities; the white
triangle is the flux tower).
![]()
EC system performance was assessed by the energy balance closure which was
calculated by conducting the linear regression between available energy
(Rn-G) and the sum of surface fluxes (λET+ H), which is also used to examine the
quality of flux data (Wilson et al., 2002). The linear regression yielded a
slope of 0.87, an intercept of -1.42 W m-2, and an R2 of
0.82. These indicators suggested that the measurements at our experimental
site provided reliable flux data and that the EC measurements underestimated
the sum of the surface fluxes to the extent of 13 %. Many researchers have
investigated energy imbalance (Barr et al., 2006; Wilson et al., 2002;
Franssen et al., 2010), and there is a consensus that it is difficult to
examine the exact reasons for the imbalance.
Characteristics of environmental variables
A brief summary of key environmental variables is presented in this section.
Four-year and long-term (1954–2014) average monthly values of
Ds, Ta, RH, and P are shown
in Fig. 4. Monthly Ds was much higher than the long-term
average monthly values, except in July and September. The highest value of
Ds was observed in May (299.5 h) and the lowest was observed
in February (206.6 h). The seasonal characteristics of Ta showed
a highly similar pattern to that of long-term average monthly values, and the
differences were less than 1 ∘C, except in July, January, and March.
The highest value of Ta was observed in July (22.1 ∘C)
and the lowest was observed in December (-8.1 ∘C). The values of
RH were almost lower than the long-term average monthly values,
especially in March and April. The highest RH was observed in
September (65.4 %) and the lowest was observed in March (35.1 %). The
seasonal distributions of P were consistent with the long-term
average monthly values, and 89.7 % of P occurred in the growing
season. P was highest in July (120.5 mm) and lowest in January
(0.3 mm).
Seasonal characteristics of 4-year and long-term (1954–2014, from
Yulin meteorological station) average monthly values of (a) sunshine
duration (DS); (b) air temperature
(Ta); (c) relative humidity (RH);
and (d) total precipitation (P).
![]()
The inter-annual characteristics of daily Ta, Ds,
RH, θr, groundwater level (GWL), and total
P in the growing season of each period are listed in Table 1.
Daily air temperature (Ta, ∘C),
relatively humidity (RH, %), total sunshine duration
(DS, h), soil water content of the root zone
(θr, m3 m-3), the groundwater level (GWL, m), and total precipitation
(P, mm) in 1954–2014 and in the growing season of each period
(because there were some missing data in Period IV (from 12 September to 23 November 2014 and from 13 March to 22 April 2015), we excluded
data in these two time ranges of Periods I–III and 1954–2014).
Variable
1954–2014
I
II
III
IV
Ta (∘C)
19.8
19.6
20.4
19.9
19.3
RH (%)
57.7
57.3
54.9
53.4
52
DS (h)
213.3
220.7
215.8
218.2
220.7
P (mm)
329.8
357.1
384.1
330.2
199.8
θr (m3 m-3)
–
0.077
0.077
0.076
0.064
GWL (m)
–
-3.8
-3.6
-3.0
-3.5
The values of Ta, RH, P, and
θr in the growing season of Period IV were the lowest
compared to those in the other three periods. Periods I–III were all
wet years, while Period IV was a dry year. The values of θr
in Periods I–III were similar; however, θr decreased by
0.0113 m3 m-3 in Period IV. The mean GWL in Period III was the
shallowest.
Seasonal variations in <italic>E</italic><inline-formula><mml:math id="M278" display="inline"><mml:msub><mml:mi/><mml:mi mathvariant="normal">T</mml:mi></mml:msub></mml:math></inline-formula> due to climate variability and
vegetation phenology
The seasonal curve of ET in each year had a single peak
value (Fig. 5a), with higher ET appearing mostly in the
growing season, while lower ET appeared in the
non-growing season. The daily ET ranged from
0.0 mm day-1 to 6.8 mm day-1 during the four periods; the
highest ET was observed on 22 June 2013, which was the
day after a continuous rainfall event that extended from 19 to 21 June 2013
(90.3 mm). The lowest ET appeared on 28 November 2012,
which was in the frozen period (late November to early March at our study
site). On rainy days, ETP (Fig. 5b) was low due to low
net radiation and air temperature. ETP ranged from
0.2 mm day-1 in December 2011 to 17.9 mm day-1 in
September 2013.
Seasonal and inter-annual characteristics of
daily (a) evapotranspiration
(ET, mm); (b) potential evapotranspiration
(ETP, mm); (c) NDVI in zone A and zone B within
the footprint; (d) the soil water stress of the root zone
(fs); and (e) the groundwater level (GWL, m) from
1 July 2011 to 30 June 2015.
![]()
The seasonal NDVI curve for natural land use/cover conditions (in zone A
during Periods I–IV and in zone B during Period I) represented the process
of natural vegetation phenology, and it had a single peak value in each year
(Fig. 5c). In early May, the seasonal NDVI curve began to increase as the
native vegetation entered the growing season, and a maximum value
(0.27 ± 0.01) was reached in July or August. In the winter, the
daily NDVI remained relatively constant (0.13 ± 0.01). fs
(Fig. 5d) increased rapidly in response to rainfall events of more than 5 mm
a day and decreased rapidly 1 or 2 days after rainfall events. From late
November to early March, there was a frozen period when the soil water
content was below the wilting point. The groundwater level changed obviously
in the monsoon season (July to September) and mildly in the winter (December
to February).
The correlations between daily evapotranspiration
(ET, mm) and its controlling factors: (a) daily
potential evapotranspiration
(ETP, mm); (b) daily weight-averaged NDVI
(NDVIw) within the footprint; and (c) daily soil
water stress of the root zone (fs) in Period I by excluding the
data on rainy days (r: Pearson's correlation coefficient; T: t test
significance).
![]()
The linear correlations between ET and the three controlling
factors all passed the t test at a 95 % confidence level. The R2
value of the correlation between ET and
NDVIw
(NDVIw=NDVIA×β+NDVIB×(1-β)) was the largest,
indicating that NDVI was highly correlated with the daily variations in
ET. To better quantify the effects of the phenological process
on ET, the correlation between daily fv and
NDVIw in Period I was analyzed (Fig. 7a).
Quantitative analysis of the correlations
between (a) vegetation phenological change (NDVIw)
and daily normalized ET (fv=ET/(ETP×fs)) in Period I (excluding the data
on rainy days and frozen days) and (b) the indicator of land
use/cover change (Dlu) and total normalized
ET (fv=ET/(ETP×fs)) in the growing season of each
period.
![]()
A positive linear regression was found between fv and
NDVIw (Fig. 7a). The slope of the linear regression
was used to evaluate the degree of the correlation between fv
and the vegetation phenological process. We found that when
NDVIw increased by 1 unit, fv
increased by approximately 1.86 units.
Inter-annual variations in <italic>E</italic><inline-formula><mml:math id="M325" display="inline"><mml:msub><mml:mi/><mml:mi mathvariant="normal">T</mml:mi></mml:msub></mml:math></inline-formula> due to land use/cover
change
During the four periods, in zone A, the NDVI values of each period were
similar because the land use/cover condition did not change, while in zone B,
the peak values of NDVI first declined from 0.28 to 0.15 (Period I to
Period III) due to the land use/cover condition changing from mixed
vegetation to bare soil. The peak NDVI values then increased to 0.22
(Period IV) due to grass recovery (Fig. 5c). An interesting phenomenon was
observed accompanied by the changing process of land use/cover conditions:
ET in the growing season gradually increased from
Period I to Period III (Table 2), while it increased greatly in Period IV
even with less precipitation, because a mass of soil water and groundwater
was consumed to satisfy the ET demand (Fig. 5e).
Typical values of total evapotranspiration
(ET, mm), total potential evapotranspiration
(ETP, mm), the indicator of land use/cover change
(Dlu, %), the soil water stress of the root zone
(fs), and normalized ET (fv
(=ET/(ETP×fs))) in the growing
season of each period (because there were some missing data in Period IV
(from 12 September to 23 November 2014 and from 13 March to 22 April 2015),
we removed the values of ET, ETP,
and fs in these two time ranges of Periods I–III).
Period
ET (mm)
ETP (mm)
Dlu (%)
fs (dimensionless)
fv (dimensionless)
Growing season
I
238.4
876.1
-0.2
0.62
78.1
II
236.5
870.7
4.6
0.63
79.9
III
292.1
956
10.4
0.59
86.3
IV
332.2
937
6
0.37
111.9
Compared with Period I, Dlu values in Period II and Period III
gradually increased, while Dlu in Period IV decreased.
Taking August in each period as an example, in Period I,
Dlu was 0.2 %, while in Periods II–IV,
Dlu were 2.9, 12.6, and 8.6 %, respectively. In
order to eliminate the influence of vegetation phenological change
on ET, we chose the growing season of each period to analyze
the correlation between fv and Dlu.
The quantitative results of the correlation between Dlu and
fv are shown in Fig. 7b. From Period I to Period III, as land
surface characteristics changed (the natural vegetation in zone B was
cleared, the fixed and semi-fixed sand dunes were bulldozed, and the BSCs and
dry sand layers disappeared), fv increased, and this increase
was more evident in Period III (from 78.5 to 88.1). When the land use/cover
conditions in zone B gradually changed from bare soil to sparse grassland due
to the self-restoring capacity of nature, fv increased
significantly (from 88.1 to 111.3).
Discussion
Implications of the effects of phenological change on
<italic>E</italic><inline-formula><mml:math id="M355" display="inline"><mml:msub><mml:mi/><mml:mi mathvariant="normal">T</mml:mi></mml:msub></mml:math></inline-formula>
The correlations between ET and its controlling factors
suggest that at our experimental site, NDVI is the predominant
factor that influences the seasonal variations in ET. The
positive linear relationship between fv and NDVI suggests that
transpiration is likely controlled by the stomatal conductance and the
numbers of stomata, which are proportional to the leaf area (Pearcy et al.,
1989; Turrell, 1947), rather than the atmospheric water demand represented
by ETP.
Various studies have assessed the correlation between vegetation phenological
change and ET, and these results generally reflected
consistent and positive linear relationships (Nouri et al., 2014; Rossato et
al., 2005; Duchemin et al., 2006; Glenn et al., 2007). However, for different
vegetation species, phenological change has effects on
ET to different degrees. Relatively strong regressions
between NDVI and ET have been reported at forested sites
(Loukas et al., 2005; Nouri et al., 2014; Lo Seen Chong et al., 1993) and
grass-covered sites (Kondoh and Higuchi, 2001; Nouri et al., 2014), with
determination coefficients higher than 0.7. These results reflect the strong
control between phenological changes and ET. Thus, we
speculate that for high vegetated ecosystems, phenological change may have a
significant control on ET. However, in low vegetated
ecosystems such as the sparse shrubland in this study, the relationship
between ET and phenological change is thus positive but
relatively weak.
Possible reasons for the effects of land use/cover changes
During Periods I–IV, the land use/cover conditions at our experimental site
underwent changes associated with two processes: land degradation process
(Periods II–III) and vegetation rehabilitation process (Period IV). Notable
results were observed during these two processes: (1) ET and
normalized ET values both increased and (2) normalized
ET increased much faster during the vegetation rehabilitation
process than it did during the land degradation process.
The effect of phenological change on ET demonstrates
that ET decreases with leaf browning. Thus, we expect
that ET will also decrease if leaves are cleared by
human activities. However, during Periods II–III, not only were leaves
cleared, but other land surface properties were also changed (all branches
were cut, sand dunes (fixed and semi-fixed) were bulldozed, and the dry sand
layers and BSCs were destroyed), resulting in complex land use/cover
conditions. These altered land surface properties might contribute to the
increase in ET. Previous studies demonstrated that dry
sand layers and BSCs could effectively restrict the soil evaporation rate (Z.
Wang et al., 2006; Lv et al., 2006; Liu et al., 2006; Dong et al., 1999;
Yang et al., 2015; Fu et al., 2010; Liu, 2012). However, the bulldozing of
sand dunes at our experimental site made the elevation of the flat soil
surface lower than the average elevation of the undisturbed soil surface
(approximately 1.5 m lower, Fig. 2d), making the groundwater depth much
shallower than the pre-disturbance depth. Thus, the formation of dry sand
layers was restricted due to the shallow groundwater level. In this situation
with the destroyed BSCs and the disappeared dry sand layers, the sufficient
groundwater supply (Li and Li, 2000) accelerated the loss of water that was
stored in shallow soil through evaporation. The enhanced soil evaporation
offset the inhibiting effect of transpiration due to leaves clearing, which
made ET increase.
A secondary reason for the increase in soil evaporation was that the soil
layer absorbed more solar radiation during the land degradation process. In
Period I, the radiation absorbed by the shadowed soil was the solar radiation
transmitted into the canopy of shrubs and grass. However, when the natural
vegetation was cleared, the leaves and the branches were also removed, which
made the shadowed soil exposed and enhanced the radiation absorbed by the
soil, thereby increasing soil evaporation (Martens et al., 2000; Panferov et
al., 2001). Moreover, the removal of leaves and branches and the
disappearance of sand dunes both altered the land surface albedo, which could
have directly altered the solar radiation absorbed by the land surface
(Dirmeyer and Shukla, 1994; Greene et al., 1999), subsequently leading to the
change in ET.
Some inconsistent results regarding the ET dynamics
during land degradation process were reported. A portion of studies reported
that ET decreased during the land degradation process
for different reasons. For example, Souza and Oyama (2011) and Snyman (2001)
demonstrated that ET decreased during the land
degradation process due to decreased transpiration in semiarid regions. Lu et
al. (2011) considered that the low soil water content was the main reason for
the decrease in ET during the land degradation process.
Mao and Cherkauer (2009) also reported a decrease in ET
when the land use/cover condition was converted from forest to grass or
cropland in the Great Lakes region. However, contrasting results were also
reported regarding the effects of land degradation on
ET. Hoshino et al. (2009) found that there was no
difference in ET during the land degradation process
associated with overgrazing in a semiarid Mongolian grassland, and they
hypothesized that the reason for this lack of change might be the short
grazing time (2 years). Li et al. (2013) demonstrated that the warming air
temperature was the main cause of increased ET during
the land degradation process on the Qinghai–Tibet Plateau. Throughout the
above studies of ET during land degradation processes,
we found it difficult to accurately describe the trends in
ET, even when the land degradation was only manifested
by less vegetation coverage. Therefore, at our study site with complex land
surface properties (sand dunes, dry sand layers, and BSCs), the effect of
land degradation on ET was much more complicated.
During the vegetation rehabilitation process (Period IV), fv
increased significantly due to the rehabilitation of grass in zone B, even
though less precipitation was observed compared with other periods
(Periods I, II, and III). The rehabilitation of grass, rather than shrubs,
was due to the sufficient groundwater supply, which resulted from bulldozing
the sand dunes. Previous researchers reported that sparse shrubs more
commonly grew at the top of sand dunes and that grass grew at the bottom of
sand dunes because the difference between groundwater level and the top of
sand dunes was larger than that between groundwater level and the bottom of
the sand dunes (Lv et al., 2006; Dong et al., 1999). Because transpiration
increases with vegetation greening (as demonstrated in Sect. 4.3), the
regrowing grass would enhance plant transpiration supplied by the sufficient
groundwater. More importantly, the transpiration rate of grass is higher than
that of shrubs because shrubs are easier to survive in water-limited
conditions (Yang et al., 2014; Wang et al., 2002; Wu, 2006). Therefore, in
the vegetation rehabilitation process, the enhancement of the transpiration
rate in Period IV was much higher than that in Periods I–III. Similar
conclusions regarding increased ET due to the enhanced
transpiration during the vegetation rehabilitation process were reported (Qiu
et al., 2011; Yang et al., 2014; Sun et al., 2006; Li et al., 2009).
Meanwhile, the regrowing grass could reduce the radiation absorbed by the
soil and hence reduce soil evaporation. However, the interception of
radiation by the grass canopy was expected to be smaller than that by the
mixed shrub and grass canopy in Periods I–III because the leaf area index of
grass was smaller than the sum of leaf area and stem area indexes of the mix
of shrubs and grass. Therefore, the reduction in soil evaporation in
Period IV might be small compared with the increase in soil evaporation in
Periods I–III.
We noticed that the GWL decreased continuously from Period III to Period IV
due to the enhanced ET by the regrowth of grass and relative
low precipitation, and the regrowing grass has a higher transpiration rate
than that of the native mixed shrub and grass. Therefore, we hypothesize
that if the land use/cover condition of zone B continues to be grassland
over the next several years, the groundwater level will decrease due to the
larger consumption, making the soil water condition gradually become poorer
for the growth of grass. Then, in this situation, the grassland is expected
to degrade to shrubland in zone B because shrubs are easier to survive in
water-limited ecosystems. Furthermore, in the next few years, potatoes will
be planted in zone B. However, the water requirement of potato is more than
320 mm in the growing season (Qin et al., 2013; Liu et al., 2010) and the
water consumption is more than that of natural grass (Qin et al., 2013,
2014; Hou et al., 2010). Thus, irrigation is necessary for planting potatoes
during the growing season in water-limited ecosystems (Liu et al., 2010; Fabeiro et al., 2001). Our results imply that the
groundwater level might continue to decrease faster with the growth of
potatoes in the future, which may lead to a more fragile ecosystem.