Separating the impacts of climate change and human activity on actual evapotranspiration (ET) is important for reducing comprehensive risk and improving the adaptability of water resource systems. In this study, the spatiotemporal distribution of actual ET in the Aksu River Basin, Northwest China, during the period 2000–2015 was evaluated using the Vegetation Interfaces Processes model and Moderate Resolution Imaging Spectroradiometer-Normalized Difference Vegetation Index. The impact of climate change and human activity on actual ET were separated and quantified. The results demonstrated that: (1) the annual pattern of actual ET per pixel exhibited the highest values for arable land (average 362.4 mm/a/pixel), followed by forest land and grassland (average of 159.6 and 142.8 mm/a/pixel, respectively). Significant increasing linear trends (p < 0.05) of 3.2 and 1.8 mm/a were detected in the arable land and forest land time series, respectively; (2) precipitation was the most significant of the selected climate factors (precipitation, average temperature, sunshine duration, and wind speed) for all ecosystems. The second most significant was wind speed; (3) human activity caused 89%, 98%, and 80% of the changes in actual ET of forest, grass, and arable land, respectively, while climate change caused 11%, 2%, and 20% of the changes in actual ET, in the Aksu River Basin during 2000–2015.
Evapotranspiration (ET) is a significant hydrological factor associated with energy balance and water budget that represents water consumption (Liu et al. 2016; Chen et al. 2017), but is difficult to evaluate at the catchment scale (Liu et al. 2016). The accurate assessment of ET plays an important role in effective water resource management (Yeh et al. 1998) and in the quantification of water resource carrying capacity (Loucks 2000). Generally, ET estimation at the basin scale is based on land surface hydrologic models combined with weather data (e.g., Wood et al. 1992), land-atmosphere models with reanalysis, or remote sensing (e.g., Rodell et al. 2004; Mu et al. 2007). Although commonly used and well-established models have typically been verified by flux tower ET observations (e.g., Running et al. 1999; Xu et al. 2006), results often vary across multiple spatiotemporal scales (Liu et al. 2016). In this study, the Vegetation Interfaces Processes (VIP) model established by Mo & Liu (2001) and developed by Mo et al. (2011) and Mo et al. (2015) was applied to estimate actual ET at the pixel scale of remote sensing imagery, with the incorporation of ground surface properties. The model involves multiple modules that simulate energy, water, and carbon exchanges between terrestrial ecosystems and the atmosphere (Mo et al. 2014). Furthermore, the model has been widely applied in China, including the Lushi Basin (Mo et al. 2004), Xitiao River catchment (Mo & Meng 2011), Northeast China (Mo et al. 2014), and North China (Mo et al. 2017). These studies demonstrate the broad applicability of the VIP model.
Climate factors and human activity greatly impact ET (Qiu et al. 2008; Liu et al. 2010; Zhang et al. 2011, 2013; Bai et al. 2014; Li 2014; Sun et al. 2016). Shi et al. (2013) concluded that climate change played the greatest role in changes to global ET during 1982–2008. However, Bai et al. (2014) found that the contribution of climate change to the ET trend was weak in the arid region of Northwest China, with relatively strong human activity during 1971–2006 (Bai et al. 2014). In addition, separating the contribution of climate change from that of human activity (i.e., irrigation, fertilization, grazing, management practices, and land use changes) is difficult (Lobell et al. 2011; Wang et al. 2016) yet important for understanding the impact of climate change on water consumption, carbon sequestration, and adaptation strategies (Chen et al. 2017). Multiple prior studies have explored the separation and quantification of the contributions of climate change and human activity. For example, Qiu et al. (2008) evaluated the ET of winter wheat and its response to different irrigation regimes in the North China Plain. Bai et al. (2014) determined that an increase in cropland area explained 60.5% of the increased total ET, and management practices accounted for 16.8% of the increased total ET. Chen et al. (2017) concluded that climate change and human activity in the North China Plain contributed 0.188 and 0.466 mm/a, respectively, to the actual ET trend of 0.654 mm/a.
The Aksu River is located in the arid northwest region of China, and nearly half of its water originates as glacier and snow melt from the mountains (Li et al. 2016). In recent years, hydrological processes in the basin have been severely affected by climate change. Jiang et al. (2005) reported that temperature increases had a greater effect on runoff than precipitation in the Aksu River. Xu et al. (2011) determined that a close relationship exists between variations in annual runoff and regional climate change. Li et al. (2016) reported that the contributions of climate factors and human activity to the runoff in the Aksu River Basin were 94% and 6%, respectively. However, few studies have attempted to separate and quantify the contributions of climate change and human activity to actual ET in the Aksu River Basin, while simultaneously considering land use changes.
The objectives of this study were to (1) evaluate actual ET in the Aksu River Basin via the VIP model and the Moderate Resolution Imaging Spectroradiometer-Normalized Difference Vegetation Index (MODIS-NDVI); (2) estimate the accuracy of actual ET in the basin based on the VIP model; and (3) quantify the contributions of climate change and human activity to actual ET in the Aksu River Basin.
STUDY AREA AND MATERIALS
The Aksu River Basin is in the northern Tarim River Basin, Northwest China (Li et al. 2016) (Figure 1). The Aksu River is one of the largest tributaries of the Tarim River by runoff volume and covers an area of 5.14 × 104 km2 (Wang et al. 2010). The basin experiences a typical temperate continental climate due to the blocking effect of the terrain and a large distance to the ocean. Thus, a low average annual precipitation (64 mm), high average annual potential ET (1,890 mm), and average annual temperature range (9.2–11.5 °C) are observed in the basin (Xu et al. 2011). In addition, basin runoff is primarily derived from the Tienshan Mountains and is influenced by the complex climatic conditions and hydrological environment (Xu et al. 2011).
Agricultural development in the Aksu River Basin is dominated by oasis irrigation agriculture, for which the irrigation water originates from the river or reservoirs (Huang et al. 2015). Tarim irrigation districts are primarily located in the downstream reaches of the Aksu River Basin, where the highest water consumption occurs (Han et al. 2015). Moreover, polluted water in the basin is re-routed for irrigation and vegetable production, causing vegetation degradation and salinization of cultivated land (Ji et al. 2000; Guo et al. 2003). The difference between the supply and demand for water in the Aksu River Basin is becoming more intense with continued economic development.
Meteorological data, land use cover change (LUCC) data, digital elevation model data, and MODIS-NDVI were used to drive the VIP model in this study. Meanwhile, hydrometeorological data were applied to construct the water balance (WB) model. Then, a WB-ET model was obtained to verify the results of the VIP model simulation.
Weather and hydrological data
Daily weather data (precipitation (P), minimum temperature (Tmn), maximum temperature (Tmx), relative humidity (RH), sunshine duration (SD), wind speed (WS), and atmospheric pressure (AP)) from 2000–2015 at three weather stations were used to drive the VIP model. These data were obtained from the China Meteorological Administration (CMA) (http://data.cma.cn/), as shown in Table 1. The station data were interpolated to the entire basin based on the gradient plus inverse distance squared method proposed by Nalder & Wein (1998). Additionally, runoff data from 2000–2011 were collected from the Annals of Hydrological Statistics, compiled by the Ministry of Water Resources of the People's Republic of China. The monthly averaged runoff was used as hydrological data in this study. The hydrological stations are listed in Table 1.
|Hydrological Stations||Long.(E)||Lat.(N)||Weather Station||Long.(E)||Lat.(N)|
|Hydrological Stations||Long.(E)||Lat.(N)||Weather Station||Long.(E)||Lat.(N)|
NDVI data for the period of 2000–2015, with a 16-day temporal resolution and 1-km2 spatial resolution, was downloaded from the MODIS product website (https://lpdaac.usgs.gov/). Missing data at the beginning of 2000 was substituted with that of 2001. NDVI data quality was then improved by application of the Savitzky–Golay (SG) filter (Savitzky & Golay 1964). A smoothing effect was achieved by fitting successive subsets of adjacent data points with a low-order polynomial. The filter can reduce or eliminate the impact of clouds on the quality of remote sensing images. Finally, the 16-day NDVI data was interpolated to daily data using the Lagrange polynomial method to match the temporal resolution of the VIP model.
(LUCC data for 2000, 2005, and 2010 were provided by the Data Center for Resources and Environmental Sciences, Chinese Academy of Sciences (RESDC) (http://www.resdc.cn). The dataset is primarily based on Landsat Thematic Mapper/Enhanced Thematic Mapper remote sensing images and an artificial visual interpretation method (Liu et al. 2014). In this study, 1-km2 resolution LUCC data were used to drive the VIP model to estimate actual ET during 2000–2002, 2003–2007, and 2008–2015.
Water balance model
For this study in the Aksu River Basin, two hydrological stations (Shaliguilanke and Xiehela) in the mountainous region were defined as inflow stations, and one station (Xidaqiao) was defined as the outflow station.
Separating the impacts of climate change and human activity
Subsequently, changes in the actual ET caused by other human activities (i.e., irrigation, fertilization, grazing, and management practices, among others) and by climate change are processed based on the methods of Nicholls (1997) and Chen et al. (2017). Due to their non-climatic influences on ET (Nicholls 1997; Lobell & Field 2007; Tao et al. 2008; Veron et al. 2015), these factors can be removed by applying a first-difference detrending method (i.e., the difference of the value between one year and the previous year) (Chen et al. 2017). The relationships between the detrended climate factors and the detrended ET are evaluated by partial correlation analysis, which can explore the relationship of two variables independent of the influences of other factors (Nicholls 1997; Xiao et al. 2015; Dass et al. 2016). The highest partial correlation coefficient indicates the most dominant climate factor (Chen et al. 2017).
RESULTS AND DISCUSSION
LUCC in the Aksu River Basin
Land use changes in the Aksu River Basin were investigated using the LUCC dataset for 2000, 2005, and 2010, as shown in Figure 2. While grassland and glaciers converge in the north of the basin, unused and desert land characterize the southwest of the basin. Grassland covers the largest area of the Aksu River Basin. Arable, residential, and industrial land are mostly located near the middle and lower reaches of the river. Arable land increased from 4,672 km2 in 2000 to 5,420 km2 in 2005, and 5,598 km2 in 2010. In contrast, forest land decreased from 1,291 km2 in 2000 to 1,213 km2 in 2005, and 1,203 km2 in 2010. Similarly, grassland decreased from 19,996 km2 in 2000, to 19,524 km2 in 2005, and 19,467 km2 in 2010. Residential and industrial land collectively increased by 5.6% from 2000 to 2005. Chen et al. (2016) observed similar LUCC patterns and reported a decreasing trend in natural land cover and an increase in artificial land cover in the Tarim River Basin.
Because MODIS-NDVI was a significant driving factor of the VIP model, it was also analyzed. The monthly spatiotemporal distribution of NDVI in the Aksu River Basin during 2000–2015 is shown in Figure 3. Figure 3(a) and 3(b) are the original NDVI, while Figure 3(c) and 3(d) are the SG-filtered NDVI. The highest NDVI values were concentrated on the river bank and arable land, and the lowest NDVI was located in the north of the basin in a glaciated area. The NDVI time series exhibited an insignificant increasing trend during 2000–2015, with maximum and minimum NDVI recorded in July 2012 and January 2008, respectively. This increase can be attributed to an increase in arable land during this period (Huang et al. 2015). A significant smoothing effect can be seen in Figure 3(d), justifying the use of the filter to eliminate the effect of clouds on the remote sensing images.
Assessment of actual ET using the VIP model
To assess the accuracy of the actual ET derived from the VIP model, the driving factors of the WB model and WB-ET model were analyzed (Figure 4). Precipitation decreased from 2000–2007 but increased from 2008–2011. Meanwhile, the maximum retained runoff (i.e., Rin−Rout) occurred in 2005 and significant seasonal fluctuations are evident in both time series. While the VIP-ET series exhibited significant seasonal fluctuations, VIP-ET displayed a significant correlation with WB-ET. For instance, apart from the period 2003–2005, the amplitudes and phases of the actual ET time series were consistent, with a coefficient of determination of 0.79 at a 95% significance level and a root mean square error (RMSE) of 13.7 mm. From this result, we conclude that VIP-ET can be applied to determine the actual ET on the ground, e.g., for vegetation and arable land.
Spatiotemporal distribution of actual ET from the VIP model in the Aksu River Basin
Figure 5 shows the annual spatiotemporal distribution of actual ET derived from the VIP model, including consideration of land use changes. The pixels of actual ET were divided by ecosystem (i.e., forest land, grassland, and arable land) based on LUCC data. The annual mode of actual ET per pixel displays the highest values for arable land (average 362.4 mm/pixel), followed by forest land and grassland (average of 159.6 and 142.8 mm/pixel). In addition, in the annual time series of actual ET (Figure 5(b), 5(d), and 5(f)), the lowest value occurred in 2000 (108 mm for forest, 98 mm for grassland, and 315 mm for arable land), followed by 2009. The highest value occurred in 2013, followed by 2010 and 2003. Significant increasing linear trends (p < 0.05) of 3.17 and 1.8 mm/a were observed in arable and forest time series, respectively. As shown in Figure 2, the largest amount of water consumption in grassland can be attributed to this land use type having the largest area. In addition, the increase of actual ET in arable regions was due to the increased arable land area.
To separate the contribution of LUCC to actual ET, the VIP model was run again with climate factors and parameters identical to those used in the first process. However, the input land use was fixed at conditions recorded in 2000. The annual spatial distribution of actual ET from forest, grassland, and arable land is shown in Figure 6(a), 6(c), and 6(e). Arable land displayed the highest per pixel actual ET (367 mm/a/pixel), followed by forest and grassland (162 and 146 mm/a/pixel). In addition, the time series of actual ET (Figure 6(b), 6(d), and 6(f)) displayed significant fluctuations. The grassland and forest land time series exhibited relatively lower actual ET during 2000, 2009, and 2014–2015, while the arable time series was more stable. Meanwhile, significant increasing trends (p < 0.05) were observed in the time series of actual ET from forest and arable land, which were 2.36 and 3.17 mm/a/a, respectively. Thus, irrespective of land use changes, there was a strong similarity in the spatiotemporal distribution of actual ET derived from the VIP model. The difference in the two slopes for actual ET from the two VIP model runs was regarded as the LUCC contribution. It is notable that there was a difference of 0.5, 0.3, and −0.002 mm/a/a in the results of the VIP model for forest, grass, and arable land cover. This means that LUCC contributed 17.6%, 43.3%, and 0.08% to actual ET for forest, grass, and arable land, respectively.
Separating the impact of climate change and human activity on water consumption in the Aksu River Basin
The VIP model was driven by multiple factors: P, Ta, Tmx, Tmn, AP, RH, WS, and SD. Thus, the analysis of their impacts on ET is important to quantify the contribution of climate change to actual ET. However, Chen et al. (2017) concluded that there were significant auto-correlations among these climate factors. Thus, it was necessary to check the correlation among climate factors prior to multiple regression. The simple correlation coefficients and their significance levels for every two original climate factors and detrended climate factors for forest land, grassland, and arable land are displayed in Figure 7. Ta, Tmn, and Tmx agree well with each other at the 95% significance level, as does RH with WS and Ta with AP. Ukkola & Prentice (2013) and Cao et al. (2014) reported that SD, Ta, P, RH, and WS were the key climate factors influencing ET; thus, Tmn, Tmx, AP, and RH were not considered in the study. Subsequently, the relationships between ET and the above climate factors (i.e., P, Ta, SD, and WS) were analyzed for the Aksu River Basin.
To obtain the multiple regression models, both the climate factors and actual ET were detrended. Figure 8 depicts the percentile distribution of the original annual actual ET and detrended actual ET for different ecosystems in the Aksu River Basin during 2000–2015. Figure 8(a) shows that the actual ET of arable land is the highest, while Figure 8(b) shows that the detrended actual ET of arable land exhibits the narrowest range. This result means that the impact from climate change on actual ET was evident for arable land, but unclear for grassland. The normalized regression coefficients for the multiple linear regression model are shown in Table 2. P was most significant among the selected climate factors (P, Ta, SD, and WS) for all ecosystems, and WS was the second most important. Ta and SD displayed varying significance for different ecosystems.
Based on the method of Nicholls (1997) and Chen et al. (2017), the contribution of LUCC to actual ET was incorporated into the evaluation. The contributions of climate change and human activity to actual ET in the Aksu River Basin are shown in Table 3. The results indicate that human activity dominated actual ET changes in the Aksu River Basin, while climate change caused less significant changes. Human activity contributed 89%, 98%, and 80% to actual ET changes in forest, grass, and arable land, respectively, and climate change caused actual ET changes of 11%, 2%, and 20%, respectively, during 2000–2015. The relatively well-developed irrigation agriculture and widespread animal husbandry in the basin could result in a more significant impact of human activity on arable land and grassland, but not forest land. Bai et al. (2014) determined that this was the case in the Sangonghe Catchment, Northwest China, where human activity explained 77.3% of the changes in actual ET, whereas climate factors accounted for only 22.7%. We conclude that human activity dominates changes in actual ET in the Aksu River Basin.
The spatiotemporal distribution of actual ET in the Aksu River Basin during 2000–2015 was evaluated using the VIP model and 1-km2 MODIS-NDVI data. The actual ET derived from the VIP model was accessed by a WB model, and the contributions of climate change and human activity on the trends of actual ET were separated and quantified. The major conclusions are as follows.
Based on LUCC data, a decreasing trend was observed in forest land and grassland, but an increasing trend was observed in arable land in the Aksu River Basin. Arable land increased from 4,672 km2 in 2000 to 5,420 km2 in 2005 and 5,598 km2 in 2010. In contrast, forest land decreased from 1,291 km2 in 2000 to 1,213 km2 in 2005 and 1,203 km2 in 2010. Grassland decreased from 19,996 km2 in 2000 to 19,524 km2 in 2005 and 19,467 km2 in 2010.
The annual pattern of actual ET per pixel displayed the highest values for arable land (average 362.4 mm/a/pixel), followed by forest land and grassland (average of 159.6 and 142.8 mm/a/pixel, respectively). Significant increasing linear trends (p < 0.05) of 3.2 and 1.8 mm/a were detected in the arable land and forest land time series, respectively. In terms of actual ET derived from the VIP model with constant LUCC, arable land displayed the highest per pixel actual ET (367.2 mm/a/pixel), followed by forest and grassland (162.1 and 145.7 mm/a/pixel, respectively). Meanwhile, significant increasing trends (p < 0.05) were observed in the time series of actual ET from forest and arable land, which were 2.4 and 3.2 mm/a, respectively.
Ta, Tmn, and Tmx agreed well with each other at a 95% significance level, as did RH with WS, and Ta with atmospheric pressure. Precipitation was the most important factor affecting the selected climate factors (P, Ta, SD, WS) for all ecosystems, and the second most important factor was wind speed.
Human activity contributed 89%, 98%, and 80% to actual ET changes in forest land, grassland, and arable land, respectively, while climate change contributed to actual ET changes of 11%, 2%, and 20%, respectively, in the Aksu River Basin during 2000–2015.
This research is supported by the National Basic Research Program of China (973 Program, No. 2015CB452701) and the National Natural Science Foundation of China (No. 41571019). The first author thanks the National Aeronautics and Space Administration (NASA) for providing the MODIS data used in this study.