Spatiotemporal Changes in Evapotranspiration from an Overexploited Water Resources Basin in Arid Northern China and Their Implications for Ecosystem Management

Evapotranspiration (ET), including evaporation from soil and water surfaces and transpiration from vegetation, influences water distribution in the soil-plant-atmosphere continuum, especially in arid areas where water is a key limiting factor. Therefore, understanding the spatiotemporal dynamics of ET, including its two components of soil evaporation (Es) and vegetation transpiration (Ec), can be useful for water resource management and ecological restoration in arid regions. Based on ET data from 2002 to 2012, the spatiotemporal variations in ET were evaluated in the Shiyang River Basin in arid Northwest China. The results showed the following: (1) spatially, ET decreased from upstream of the Qilian Mountains to the middle and downstream, with a mean annual value of 316 mm; (2) temporally, ET showed a single peak curve throughout the year, with the highest value occurring in summer; (3) ET showed a downward trend (from 350 to 265 mm) before 2009 and thereafter increased (from 265 to 345 mm); and (4) water use efficiency, indicated by the ratio of Ec to ET, was low in the cropland, with a mean value of 50.9%. Further analysis indicates that decreases in ET are mainly caused by vegetation decreases; in contrast, ecological restriction measures and strict water resource management policies in the middle reaches of the basin led to ET increases. It is concluded that understanding ET and its two components can elucidate the connections between water and human society.


Introduction
The goal of sustainable development is to balance the human-nature equilibrium. However, it is challenging to achieve this goal because the human-nature system comprises complex interactions among hydrologic systems and climatic, ecological, biophysical, biochemical and socioeconomic systems [1,2]. Among these systems, the hydrologic system, including the processes of evapotranspiration (ET), condensation, precipitation, infiltration, streamflow, and groundwater Qilian Mountain area in the upper south, with an elevation of 2000-5500 m, a plain area in the middle and a hilly and desert area in the lower north, with flat terrain and an elevation of 1300-2000 m. The climate in the basin varies distinctly from south to north: the upstream Qilian Mountain area has a mean annual precipitation of 300-600 mm and a potential ET of 700-1200 mm; the Hexi Corridor in the midstream is arid cold and has a mean annual precipitation of 150-300 mm and a potential ET of 1200-2000 mm; and in the lower reaches, precipitation is less than 150 mm, while the amount of potential ET exceeds 2000 mm.

Data and Analysis Method
The datasets used for this study include data on remotely sensed ET and its components data, remotely sensed land use and vegetation index data, meteorological observations, geospatial data, and social statistics.
The long-term ET data (MOD16, version 5) from 2002 to 2012 [25], with a spatial resolution of 1 km, was provided by the National Aeronautics and Space Administration (NASA) and used for the spatiotemporal analysis. The MOD16 was estimated using the Penman-Monteith (PM) method (Eq. 1), global reanalysis meteorological data, and MODIS (the Moderate Resolution Imaging Spectroradiometer)-based remotely sensed data (version 5), including the vegetation index and land use maps [25,26]. A detailed description and parameterization of the model can be found in Mu et al. [25,26]. The 8-day MOD16A2 data were retrieved and averaged for the Shiyang River Basin and used to perform the spatial and temporal statistics, based on which the variation in ET was assessed and analyzed.
where λE is latent heat flux (λ is latent heat of vaporization and E is evapotranspiration) in W m -2 ; Rn and G are the net radiation and soil heat flux in W m -2 , respectively; s is the slope of the saturation vapor pressure curve; ρair is the air density; Cp is the specific heat of air at constant pressure; es and ea

Data and Analysis Method
The datasets used for this study include data on remotely sensed ET and its components data, remotely sensed land use and vegetation index data, meteorological observations, geospatial data, and social statistics.
The long-term ET data (MOD16, version 5) from 2002 to 2012 [25], with a spatial resolution of 1 km, was provided by the National Aeronautics and Space Administration (NASA) and used for the spatiotemporal analysis. The MOD16 was estimated using the Penman-Monteith (PM) method (Equation (1)), global reanalysis meteorological data, and MODIS (the Moderate Resolution Imaging Spectroradiometer)-based remotely sensed data (version 5), including the vegetation index and land use maps [25,26]. A detailed description and parameterization of the model can be found in Mu et al. [25,26]. The 8-day MOD16A2 data were retrieved and averaged for the Shiyang River Basin and used to perform the spatial and temporal statistics, based on which the variation in ET was assessed and analyzed. where λE is latent heat flux (λ is latent heat of vaporization and E is evapotranspiration) in W m −2 ; R n and G are the net radiation and soil heat flux in W m −2 , respectively; s is the slope of the saturation vapor pressure curve; ρ air is the air density; C p is the specific heat of air at constant pressure; e s and e a are the saturation vapor pressure and actual vapor pressure in kPa, respectively; γ is the dry and wet bulb constant; r a is the aerodynamic resistance; and r s is the surface resistance in s m −1 .
Another ET dataset (MOD3T), including its two components of Es and Ec from 2008 to 2011 [27], with the same spatiotemporal resolution as MOD16A2 was used to compensate for the shortcomings of MOD16. Specifically, MOD16 assumed that ET was zero in sparsely vegetated and barren regions and did not provide ET components, i.e., Es and Ec [25,26]. The MOD3T was estimated based on a three-temperature (3T) model (Equation (2)) and MODIS products (version 5), such as land surface albedo and temperature, vegetation indices and the fraction of absorbed photosynthetically active radiation (FPAR) [27,28]. A detailed description of the 3T model can be found in Xiong et al. [28]. The MOD3T datasets were used to perform the spatiotemporal analysis of ET in sparsely vegetated areas and to assess water use efficiency (WUE), represented by Ec/ET, for farmland.
where Es and Ec are the soil and vegetation components of ET, respectively, in mm s −1 ; T a is the air temperature in K; and T 0c and T 0s are the temperatures (K) of the soil and vegetation components, respectively. The subscripts, "s" and "c" represent the soil and canopy components, respectively, whereas "sr" and "cr" represent the reference dry soil and reference dry canopy, respectively. Daily meteorological data, including net radiation, rainfall, and air temperature, observed mainly at four national meteorological stations, namely, Wushaoling, Yongchang, Wuwei and Minqin, inside the basin from the China Meteorological Administration were averaged to represent the values of the entire basin.
Social data sets such as population and water related indices, including irrigation area, pumped groundwater, well numbers, and groundwater level were used to reflect the water consumption in the study basin in recent decades. Then, variations in these datasets were used to analyze the possible impact of human activities on the local water resources, vegetation, and ET. Figure 2 shows the spatial distribution of the annual ET for the study basin between 2002 and 2012. The MOD16 datasets assumed that ET was zero in the sparsely vegetated and barren areas [25,26], which is why the lower basin is shown in gray in Figure 2. As such, the following statistical results are based on the vegetated areas in the basin. The results show that the ET values were generally high in the southern areas (Qilian Mountains) and in the oasis, where vegetation coverage was high. The annual ET gradually declined from southwest to northeast, with the highest values in the upper Qilian Mountain area in the south (>450 mm y −1 ), followed by the middle plain area (250-400 mm y −1 ) and the lower reaches (<250 mm y −1 ). The characteristics of the ET distribution were consistent with the variation in climate and vegetation in the study region.  Figure 3 shows the interannual change in ET. There were obvious fluctuations during the study period, with mean, maximum, and minimum values of 316 mm, 265 mm, and 364 mm, respectively. The annual ET declined before 2009 (R 2 =0.72), whereas it has increased since 2009, with an average rate of 27 mm a -1 (R 2 =0.99). Figure 4 shows the intra-annual change in the multiyear average ET. Generally, the ET showed a sinusoidal variation for the multiyear period. Specifically, the ET values were lowest during winter and were high during spring because the temperature was low and the vegetation was dormant in winter, and the temperatures gradually increased and vegetation began to grow in spring. The maximum ET values occurred during summer as a result of increased temperatures and improved water conditions, i.e., approximately 80% of the annual precipitation was in summer. The ET values decreased as the temperatures dropped, and the vegetation growth slowed and ultimately stopped during autumn. The temporal ET variations were correlated with the climate of the basin.   Figure 3 shows the interannual change in ET. There were obvious fluctuations during the study period, with mean, maximum, and minimum values of 316 mm, 265 mm, and 364 mm, respectively. The annual ET declined before 2009 (R 2 = 0.72), whereas it has increased since 2009, with an average rate of 27 mm a −1 (R 2 = 0.99). Figure 4 shows the intra-annual change in the multiyear average ET. Generally, the ET showed a sinusoidal variation for the multiyear period. Specifically, the ET values were lowest during winter and were high during spring because the temperature was low and the vegetation was dormant in winter, and the temperatures gradually increased and vegetation began to grow in spring. The maximum ET values occurred during summer as a result of increased temperatures and improved water conditions, i.e., approximately 80% of the annual precipitation was in summer. The ET values decreased as the temperatures dropped, and the vegetation growth slowed and ultimately stopped during autumn. The temporal ET variations were correlated with the climate of the basin.  Figure 3 shows the interannual change in ET. There were obvious fluctuations during the study period, with mean, maximum, and minimum values of 316 mm, 265 mm, and 364 mm, respectively. The annual ET declined before 2009 (R 2 =0.72), whereas it has increased since 2009, with an average rate of 27 mm a -1 (R 2 =0.99). Figure 4 shows the intra-annual change in the multiyear average ET. Generally, the ET showed a sinusoidal variation for the multiyear period. Specifically, the ET values were lowest during winter and were high during spring because the temperature was low and the vegetation was dormant in winter, and the temperatures gradually increased and vegetation began to grow in spring. The maximum ET values occurred during summer as a result of increased temperatures and improved water conditions, i.e., approximately 80% of the annual precipitation was in summer. The ET values decreased as the temperatures dropped, and the vegetation growth slowed and ultimately stopped during autumn. The temporal ET variations were correlated with the climate of the basin.    Figure 5 evaluates the Es and Ec, the ET components of cropland during the growing season in the study basin based on the MOD3T data. Statistical results showed that the ratio of the Ec to ET (Ec/ET) ranged from 19.6 to 87.1% over the four-year period (2008-2011), with a mean value of 50.9%. In other words, approximately 49.1% of the water returned to the atmosphere in the form of soil evaporation. Although the value of Ec/ET showed some differences year to year, the change trend in each year was nearly the same, i.e., Ec/ET increased gradually before DOY 193, and thereafter, decreased gradually. Therefore, this finding indicates that certain volumes of irrigation water were lost in the form of soil evaporation.

Bias in ET Estimation
Remotely estimating ET and its components is challenging and ET estimates contain certain bias [4,29]. Nonetheless, remote sensing is the only suitable method for mapping ET across a wide spatiotemporal scale [24,29]. The adopted MOD3T data sets were previously evaluated through a Bowen ratio flux observation of a sparsely vegetated community, with a mean absolute error (mean absolute percent error) of 0.16 mm d -1 (23%) [27].
In this study and as shown in Figure 6, the ET data sets, including MOD16 and MOD3T were further assessed on a yearly scale using water balance ET, which was derived based on the water balance equation for the study basin (i.e., ET = precipitation + transferred water + irrigation water).  Figure 5 evaluates the Es and Ec, the ET components of cropland during the growing season in the study basin based on the MOD3T data. Statistical results showed that the ratio of the Ec to ET (Ec/ET) ranged from 19.6 to 87.1% over the four-year period (2008-2011), with a mean value of 50.9%. In other words, approximately 49.1% of the water returned to the atmosphere in the form of soil evaporation. Although the value of Ec/ET showed some differences year to year, the change trend in each year was nearly the same, i.e., Ec/ET increased gradually before DOY 193, and thereafter, decreased gradually. Therefore, this finding indicates that certain volumes of irrigation water were lost in the form of soil evaporation.   In other words, approximately 49.1% of the water returned to the atmosphere in the form of soil evaporation. Although the value of Ec/ET showed some differences year to year, the change trend in each year was nearly the same, i.e., Ec/ET increased gradually before DOY 193, and thereafter, decreased gradually. Therefore, this finding indicates that certain volumes of irrigation water were lost in the form of soil evaporation.  Figure 1).

Bias in ET Estimation
Remotely estimating ET and its components is challenging and ET estimates contain certain bias [4,29]. Nonetheless, remote sensing is the only suitable method for mapping ET across a wide spatiotemporal scale [24,29]. The adopted MOD3T data sets were previously evaluated through a Bowen ratio flux observation of a sparsely vegetated community, with a mean absolute error (mean absolute percent error) of 0.16 mm d -1 (23%) [27].
In this study and as shown in Figure 6, the ET data sets, including MOD16 and MOD3T were further assessed on a yearly scale using water balance ET, which was derived based on the water balance equation for the study basin (i.e., ET = precipitation + transferred water + irrigation water).  Figure 1).

Bias in ET Estimation
Remotely estimating ET and its components is challenging and ET estimates contain certain bias [4,29]. Nonetheless, remote sensing is the only suitable method for mapping ET across a wide spatiotemporal scale [24,29]. The adopted MOD3T data sets were previously evaluated through a Bowen ratio flux observation of a sparsely vegetated community, with a mean absolute error (mean absolute percent error) of 0.16 mm d −1 (23%) [27].
In this study and as shown in Figure 6, the ET data sets, including MOD16 and MOD3T were further assessed on a yearly scale using water balance ET, which was derived based on the water balance equation for the study basin (i.e., ET = precipitation + transferred water + irrigation water). Precipitation was observed and averaged from the meteorological stations shown in Figure 1, whereas the statistics for irrigation water and water transferred into the study basin were based on the Water Resources Bulletin of Gansu Province. Figure 6 illustrates that the two ET data sets were close to the water balance ET. For MOD16, the mean absolute error (MAE) between the estimated and observed ET values was 26.4 mm y −1 and the mean absolute percent error (MAPE) was 9.1%, whereas the MAE (MAPE) was 34.4 mm y −1 (11.9%) for the MOD3T. The results suggest that the ET datasets with limited error can be used to perform a reasonable analysis. Precipitation was observed and averaged from the meteorological stations shown in Figure 1, whereas the statistics for irrigation water and water transferred into the study basin were based on the Water Resources Bulletin of Gansu Province. Figure 6 illustrates that the two ET data sets were close to the water balance ET. For MOD16, the mean absolute error (MAE) between the estimated and observed ET values was 26.4 mm y -1 and the mean absolute percent error (MAPE) was 9.1%, whereas the MAE (MAPE) was 34.4 mm y -1 (11.9%) for the MOD3T. The results suggest that the ET datasets with limited error can be used to perform a reasonable analysis.

Factors Possibly Affecting ET
ET can be affected by many factors. In the study basin, ET is mainly limited by water instead of energy (e.g., [29]). The observed net radiation at the Minqin station exhibited a nonsignificant increase, with a slope of 0.14 and R 2 =0.14 ( Figure 7a). In addition, rainfall increased slightly, but the tendency was not significant (R 2 =0.11) (Figure 7b). As such, ET might be impacted by factors other than water. Considering the data availability, we mainly discuss the effects of temperature and vegetation on ET.

Factors Possibly Affecting ET
ET can be affected by many factors. In the study basin, ET is mainly limited by water instead of energy (e.g., [29]). The observed net radiation at the Minqin station exhibited a nonsignificant increase, with a slope of 0.14 and R 2 = 0.14 ( Figure 7a). In addition, rainfall increased slightly, but the tendency was not significant (R 2 = 0.11) (Figure 7b). As such, ET might be impacted by factors other than water. Considering the data availability, we mainly discuss the effects of temperature and vegetation on ET.
Theoretically, an increase in temperature is likely to cause much more water loss from evaporation and transpiration if other meteorological variables, such as precipitation, are stable. In this study, a slight increase in rainfall was observed, and both the mean annual and minimum air temperatures  Theoretically, an increase in temperature is likely to cause much more water loss from evaporation and transpiration if other meteorological variables, such as precipitation, are stable. In this study, a slight increase in rainfall was observed, and both the mean annual and minimum air temperatures exhibited an increasing trend, with R 2 values of 0.51 and 0.66, respectively (Figure 7b). However, ET showed a declining trend from 2002 to 2009. Under such conditions, the reduced ET may be caused by a decrease in vegetation transpiration due to the removal of vegetation and land desertification. Figure 8 supports the hypothesis that the change trend of the vegetation index was similar to that of ET. Specifically, the normalized vegetation index (NDVI) for the entire basin decreased between 2002 and 2008 (R 2 =0.25) and rose significantly thereafter (R 2 =0.90).   Theoretically, an increase in temperature is likely to cause much more water loss from evaporation and transpiration if other meteorological variables, such as precipitation, are stable. In this study, a slight increase in rainfall was observed, and both the mean annual and minimum air temperatures exhibited an increasing trend, with R 2 values of 0.51 and 0.66, respectively (Figure 7b). However, ET showed a declining trend from 2002 to 2009. Under such conditions, the reduced ET may be caused by a decrease in vegetation transpiration due to the removal of vegetation and land desertification. Figure 8 supports the hypothesis that the change trend of the vegetation index was similar to that of ET. Specifically, the normalized vegetation index (NDVI) for the entire basin decreased between 2002 and 2008 (R 2 =0.25) and rose significantly thereafter (R 2 =0.90).  Human activities, including intensive agricultural practices, have had many impacts on water resources and vegetation in the study basin in recent decades [15,[21][22][23][24]. Figure 9 shows changes in typical factors indicating social development. Population increased during the past 60 years, and the maximum value was almost 2.6 times that of 1949 [15,30]. Although the relative change of ET is poorly related to the change in population, farmland increased as the population increased and consumed additional water resources, especially in the 20th century. The amount of irrigated farmland increased 2.3 times that of 1949 levels, whereas the consumed amount of groundwater was 2.3 times that in 1972. This increased groundwater consumption led to an increased number of groundwater wells and overexploitation in the study basin. It was reported that 0.45 billion m 3 groundwater was overexploited and that the water resource utilization intensity was relatively high, with a value of 164% [23,31]. Because most of the water in the study basin was consumed by agriculture, ET showed a positive relation with the water consumption indices (e.g., amount of pumped groundwater and well numbers in the lower basin) during 2002-2009 ( Figure 9). These intensive activities led to land desertification and degradation, and the increased agricultural water consumption left marginal water for the ecosystem. The area of the Minqin Oasis in the lower basin has decreased approximately 300 km 2 since the 1950s.
poorly related to the change in population, farmland increased as the population increased and consumed additional water resources, especially in the 20th century. The amount of irrigated farmland increased 2.3 times that of 1949 levels, whereas the consumed amount of groundwater was 2.3 times that in 1972. This increased groundwater consumption led to an increased number of groundwater wells and overexploitation in the study basin. It was reported that 0.45 billion m 3 groundwater was overexploited and that the water resource utilization intensity was relatively high, with a value of 164% [23,31]. Because most of the water in the study basin was consumed by agriculture, ET showed a positive relation with the water consumption indices (e.g., amount of pumped groundwater and well numbers in the lower basin) during 2002-2009 ( Figure 9). These intensive activities led to land desertification and degradation, and the increased agricultural water consumption left marginal water for the ecosystem. The area of the Minqin Oasis in the lower basin tatidecreased approximately 300 km 2 since the 1950s.  Figure 8). The Chinese government has paid attention to overexploitation and related serious land desertification in the study basin. In addition to the Grain for Green Program (GGP), a nationwide environmental conservation project initiated in 1999 by the Chinese government to prevent vegetation loss [32], the local government has taken different measures to address overexploitation and ecosystem degradation. The Bureau of the Shiyang River Basin administration was founded in 2001 and is responsible for water-related problems in the basin. A special program has been launched by the bureau since 2006, including water resource allocation and protection, the prohibition of groundwater exploitation, water-saving programs, and ecosystem restoration. Noticeable evidence of vegetation recovery and ecological restoration has been observed, especially in the lower Minqin Oasis [22,24,33]. The statistics in Figure 8 suggest that the NDVI has increased since 2009, and the amount of pumped groundwater and related wells decreased in the late 2000s ( Figure 9). Because irrigation water consumption was strictly limited, the remaining water resources may be available and consumed by vegetation in the ecosystem, leading to ET increases. The results indicate that vegetation has increased since the late 2000s, and these results agree with the results of other studies (e.g., [24,34,35]). However, ET and vegetation cover showed an increasing tendency in the late 2000s (Figures 3  and 8). The Chinese government has paid attention to overexploitation and related serious land desertification in the study basin. In addition to the Grain for Green Program (GGP), a nationwide environmental conservation project initiated in 1999 by the Chinese government to prevent vegetation loss [32], the local government has taken different measures to address overexploitation and ecosystem degradation. The Bureau of the Shiyang River Basin administration was founded in 2001 and is responsible for water-related problems in the basin. A special program has been launched by the bureau since 2006, including water resource allocation and protection, the prohibition of groundwater exploitation, water-saving programs, and ecosystem restoration. Noticeable evidence of vegetation recovery and ecological restoration has been observed, especially in the lower Minqin Oasis [22,24,33]. The statistics in Figure 8 suggest that the NDVI has increased since 2009, and the amount of pumped groundwater and related wells decreased in the late 2000s ( Figure 9). Because irrigation water consumption was strictly limited, the remaining water resources may be available and consumed by vegetation in the ecosystem, leading to ET increases. The results indicate that vegetation has increased since the late 2000s, and these results agree with the results of other studies (e.g., [24,34,35]).

ET Dynamics and Their Implications for Ecosystem Management
The spatiotemporal dynamic changes of ET may be adopted as indicators of vegetation variation and water consumption. For example, in arid and semiarid regions where water resources are scarce, it is important to improve WUE by reducing soil evaporation (Es) and increasing crop transpiration (Ec) [16,22,[36][37][38][39]. Oases in the Shiyang River Basin, such as the Minqin Oasis, are the most productive agricultural areas in the Hexi Corridor. However, irrigation has consumed most of the water resources. As shown in Section 3.3, nearly half of the irrigated water was lost to the atmosphere without resulting in any productivity. From this perspective, these water resources were wasted, and it is essential to save irrigation water and improve cropland WUE in the study area. Therefore, these findings indicated that assessing the ET components, i.e., Es and Ec, might help improve WUE by reducing soil evaporation.
In addition, annual variations in ET respond quickly to soil water availability in sparsely vegetated regions [40]. Especially in the lower study basin, where the vegetation is rain fed and the ecosystem is highly vulnerable, ET provided by the MOD3T can be useful for indicating soil water and assessing vegetation conditions in the ecosystem. An example is shown in Figure 10. MOD3T can provide ET values for sparsely vegetated areas, and intra-annual variations in ET for a given region can be retrieved (Figure 10d). Intra-annual variations in precipitation observed at the Minqin station indicate that the estimated changes in the ET are reasonable. transpiration (Ec) [16,22,[36][37][38][39]. Oases in the Shiyang River Basin, such as the Minqin Oasis, are the most productive agricultural areas in the Hexi Corridor. However, irrigation has consumed most of the water resources. As shown in section 3.3, nearly half of the irrigated water was lost to the atmosphere without resulting in any productivity. From this perspective, these water resources were wasted, and it is essential to save irrigation water and improve cropland WUE in the study area. Therefore, these findings indicated that assessing the ET components, i.e., Es and Ec, might help improve WUE by reducing soil evaporation.
In addition, annual variations in ET respond quickly to soil water availability in sparsely vegetated regions [40]. Especially in the lower study basin, where the vegetation is rain fed and the ecosystem is highly vulnerable, ET provided by the MOD3T can be useful for indicating soil water and assessing vegetation conditions in the ecosystem. An example is shown in Figure 10. MOD3T can provide ET values for sparsely vegetated areas, and intra-annual variations in ET for a given region can be retrieved (Figure 10d). Intra-annual variations in precipitation observed at the Minqin station indicate that the estimated changes in the ET are reasonable. Currently, several remotely sensed global ET datasets are available, such as GLEAM (Global Land Evaporation Amsterdam Model) [41] and a global land ET [42]. If the dataset could provide ET or its components values with a spatiotemporal resolution suitable for application, it can be adopted to perform an analysis or assessment similar to that of the present study. However, due to challenges in accurately estimating remotely sensed ET [3,4,25,29], a precise and accurate ET estimation that is particularly useful for ecosystem management; such as shown in this study, the MOD3T could provide useful ET values for sparsely vegetated areas. Currently, several remotely sensed global ET datasets are available, such as GLEAM (Global Land Evaporation Amsterdam Model) [41] and a global land ET [42]. If the dataset could provide ET or its components values with a spatiotemporal resolution suitable for application, it can be adopted to perform an analysis or assessment similar to that of the present study. However, due to challenges in accurately estimating remotely sensed ET [3,4,25,29], a precise and accurate ET estimation that is particularly useful for ecosystem management; such as shown in this study, the MOD3T could provide useful ET values for sparsely vegetated areas.

Conclusions
The spatiotemporal dynamics of ET were assessed in the Shiyang River Basin in arid Northwest China from 2002 to 2012. The results showed that ET decreased from upstream of the Qilian Mountains to the middle and downstream, and that temporally, ET showed a downward trend before 2009 and thereafter it increased. Further analysis indicates that vegetation loss led to a decrease in ET; in contrast, ecological restriction measures and strict water resource management policies in the middle reaches of the basin led to an increase in ET. It is concluded that the dynamic spatiotemporal changes in ET and its two components may be adopted as an indicator of vegetation variation and water consumption.