Human activities rather than climate dominated the grassland net primary production in the three basins of the Hexi Corridor, North‐Western China

Actual net primary production (ANPP) is affected by both natural conditions and human activities in many areas, while the contributions of these factors to NPP remain unknown because it is difficult to isolate their effects. This knowledge gap hinders our implementation of environmental protection and restoration. In this study, we selected three basins of the Hexi Corridor, North‐western China, as the study area. Based on the monthly growth of grassland and field observations, we identified the undisturbed grids and then calculated the potential net primary productivity (PNPP) of these grids. Using the Carnegie‐Ames‐Stanford‐Approach (CASA) model, we calculated the ANPP in these areas. The influence of human activities on NPP (HNPP) was then calculated as the difference between ANPP and PNPP. The results showed that from 2001 to 2017, the mean grassland HNPP was −236.25 gC m−2 a−1. The area with a relative contribution index (RCI) of human activity greater than 0.5 accounted for more than 50% of the total, indicating that human activities were a dominant factor in the NPP change. After the implementation of ecological governance projects in the Hexi Corridor in 2006, HNPP exhibited a negative weakening rate of 2.79 gC m−2 a−1 (p < 0.05), and the RCI decreased from 0.708 at the beginning of the study period to 0.522 at the end (p < 0.01), indicating that reducing human interference with nature was an effective measure for ecological restoration. Based on these findings, we concluded that the changes in NPP in the three basins of the Hexi Corridor was dominated by human activities. Our results suggest that future studies pertaining to the response of NPP to climate change should consider the impact of human activities in semi‐arid regions.


| INTRODUCTION
Net primary productivity (NPP) is affected by both nature and human activities (Field et al., 1995;Raupach et al., 2008). During the past decades, the impact of human activities has increased greatly, significantly affecting the NPP in some areas (Harberl, 1997;Leemans & Zuidema, 1995). For example, it has been suggested that 69% of the NPP decrease was caused by human impacts in Northwest China (Zhou et al., 2015). Therefore, the calculation of the anthropogenic effects on NPP is a critical step toward understanding the mechanisms of ecosystem response to climate change (Eisfelder et al., 2014). However, due to the variety of human activities and substantial heterogeneity of land cover types, topography, soil, and climate conditions, the evaluation of the impact of human activities on NPP remains challenging in many ecosystems .
There are several methods to quantitatively calculate the impact of human activities on NPP. Different land cover types (Saikku et al., 2015) or the human activity index can be used to calculate the sum of land use changes to obtain the anthropogenic effects on the NPP (Razali et al., 2015). The dynamic vegetation model has been used to evaluate the influence of human activities on the parameters of vegetation growth, which could then be used to quantify the effects (Ren et al., 2012). Those models require a large amount of measured data in the whole region, or it is difficult to calculate the human impacts in the area where the land use type has not changed, or the model parameters are too many to obtain. Recently, the difference between the actual NPP (ANPP) and potential NPP (PNPP) has been widely used to evaluate the effects of human activity on NPP (Chen et al., 2017;Evans & Geerken, 2004;. This method is not only simple in principle and less in parameters, but also can estimate the indirect influence of human activities. For this method, ANPP is estimated using remote sensing data, while the estimation methods of PNPP have varied although most are based on climate models, such as the Thornthwaite Memorial (Thornthwaite, 1948) and Miami models (Lieth, 1975). Because the PNPP values from these models are difficult to validate, it is necessary to evaluate PNPP based on natural areas without human activities. On the basis of previous studies, we selected the undisturbed grids by analyzing the change characteristics of actual NPP and climate factor according to the vegetation type and human activity mode, and then estimated the potential NPP based on those grids. This method not only follows the natural law, but also can be used as a supplementary method for estimating potential NPP.
The Hexi Corridor, located in Northwest China, is a typical arid and semi-arid region with a fragile ecological environment. Since the 1960s, the population and the area of cities, towns, and cultivated land have increased, while the area of grassland and forest land decreased greatly (Wang et al., 2002;Zhou et al., 2013). Meanwhile, water consumption by industry, agriculture, and families has increased dramatically. Because of the limited surface water resources, a considerable amount of groundwater has been extracted (F. Li et al., 2013a). Under increasing environmental pressures, the local government has carried out several ecological management projects, including the Grain for Green Program, the Grazing Withdrawal Program, the Natural Forest Conservation Program, Ecological Migration, and closing wells, since 2006. It has been suggested that these projects have promoted environmental recovery (Deng et al., 2013;Zhu & Li, 2014). However, the effects of human activities on the ANPP in this area remains unknown because human activity can affect the environment positively or negatively.
In this study, we selected this arid and semi-arid region as the study area, focusing on the grassland because it is sensitive to climate change and human disturbance O'Mara, 2012). The NPP influenced by humans was determined using the difference between the ANPP and PNPP based on remote sensing data. The main goal of this study was to quantify the effects of climate change and human activity on the NPP in arid and semi-arid regions, providing a scientific basis for ecological protection and construction.
2 | MATERIALS AND DATA

| Study area
The Hexi Corridor, located in Northwest China, is an important corridor connecting Xinjiang with Central and Western Asia (Guan et al., 2018). The corridor passes through three inland river basins from east to west: the Shiyang River basin, the Heihe River basin, and the Shule River basin. From south to north, the area is composed of three subsystems: mountain, oasis, and desert ( Figure 1). Since the 1960s, the environment has experienced serious land degradation and environment deterioration and has been greatly affected by human activities (Han et al., 2014).

| Data and processing
The MOD13Q1 data (normalized difference vegetation index [NDVI], 250 m, 16 days) from 2001 to 2017 were obtained from the Level 1 and Atmosphere Archive and Distribution System (LAADS) website (https://ladsweb. modaps.eosdis.nasa.gov/). For the extraction of vegetation types, we selected images of Landsat TM and Landsat 8 from June to October 2001 with a spatial resolution of 30 × 30 m, which were derived from the Geospatial Data Cloud (http://www.gscloud.cn/).

| Calculation of ANPP
In this study, the ANPP was calculated using the Carnegie-Ames-Stanford-Approach (CASA) model (Field et al., 1995;Piao et al., 2001;Potter et al., 1993), which is the most representative light-use efficiency model and only requires a few parameters. Using remote sensing data, this model WANG and DOU | 211 can estimate NPP for long periods and over large scales with high accuracy (Z. Liu et al., 2017;Raza & Mahmood, 2018;Yu et al., 2009). In this model, the NPP is primarily determined by photosynthetically active radiation (APAR) and light-use efficiency (ε). The calculation is: where APAR(x,t) represents the photosynthetically active radiation (MJ m −2 ) absorbed by pixel × in month t, and ε (x,t) is the actual light-use efficiency (g C MJ −1 ) of pixel × in month t. APAR(x,t) and ε(x,t) are calculated as: where FPAR(x,t) indicates the absorption coefficient of the vegetation layer on incident photosynthetically active radiation, which is determined by the NDVI; SOL(x,t) corresponds to the total solar radiation (MJ m −2 ) of pixel × in month t; 0.5 indicates the ratio of photosynthetically active radiation that can be absorbed by vegetation to the total solar radiation; T ε1 (x,t) and T ε2 (x,t) indicate the stress function of the low temperature and high temperature to light-use efficiency, respectively; W ε (x,t) is the water stress coefficient, reflecting the influence of the water conditions; ε max is the maximum light-use efficiency under the optimal conditions. The actual regional evapotranspiration model (Zhou & Zhang, 1996) was used to calculate the actual evapotranspiration during the calculation of the light-use efficiency. The potential regional evapotranspiration was calculated according to the Holdridge vegetation-climate systems (Holdridge, 1957;Zhang et al., 1993). The monthly maximum light-use efficiency refers to the results proposed in previous literature (Shi et al., 2016). The calculation method of FPAR(x,t) refers to the NDVI-FPAR table provided in the NASA-MOD15 algorithm. The algorithm is expressed as follows:

| Selection of grids undisturbed by humans
PNPP is the NPP under natural conditions, that is, the NPP in areas without human activities. Theoretically, if the grids without human activities can be detected, then it would be possible to upscale the ANPP values in these grids to the entire region to obtain the PNPP (Wang, 2020). We hypothesized that the undisturbed grids are those only affected by climate under natural conditions, and thus the interannual variation should be consistent with climate conditions (C. Li et al., 2021). Because only temperature and precipitation are used in the PNPP estimation by climate models, it is reasonable to use the change in NPP to represent climate change. The Thornthwaite Memorial model (Lieth, 1975;Running et al., 2004;Thornthwaite, 1948) was applied to estimate the climate NPP, which was called PNPP_T. The calculation formulas are as follows: where PNPP_T is the annual NPP (g Cm −2 a −1 ); V is the average annual actual evaporation (mm); r is the annual total precipitation (mm); L is the annual evapotranspiration (mm); and t is the annual average temperature (°C). Based on above descriptions, the annual ANPP was compared with the PNPP_T as follows: where CR y is the annual coincidence ratio of climate and NPP during the study period; i represents the year; CR i denotes whether the ith year coincides; and n is the total number of years in the study period. Because there were statistical data errors, a threshold value was set for the determinations. If a value was larger than the threshold value, the ANPP was consistent with climate change, and the grid was regarded as the grid without human activities; otherwise, the grid was affected by human activities.
The above determination was based on an interannual scale, while it failed to detect the monthly impacts of human activities. For example, after grazing, the grassland will recover over time, and the coincidence of interannual variations would not be recognized. According to the growth of the grassland, the monthly NPP variation shows a single peak distribution, with the maximum NPP typically occurring in June or July in Hexi Corridor. The monthly NPP variation will be inconsistent with this distribution if the grassland is affected by human activity. Because it is difficult to determine whether the maximum value of NPP occurs in June or July, these months were not included in the calculation. The calculation formula are as follows: , then A ij = 1; otherwise A ij = 0, (j is January, February, March, April).
where ANPP ij is the ANPP of month j in year i; i represents the year (2001-2017); j represents the month; and R i represents whether the change in monthly ANPP in year i shows the single peak distribution. We examined if the NPP followed a peak distribution by calculating where R m represents the ratio of the monthly NPP that follows a single peak distribution and n is the total number of study years. When R m = 1, the distribution of the monthly NPP followed a single peak distribution during the 17-year period, and 0 indicates a nonsingle peak distribution.
Because there were errors during the statistical analysis, the noncoincidence of 2 months during the 17 years was allowed. The monthly NPP variation was in accordance with the single peak distribution, which is typical of grass growth. After extracting the grids of (CR y > 0.8) ∩ (R m = 1) and excluding the grids that did not meet the requirements during the field survey, 80% of the undisturbed grids were randomly selected for the simulation, and the remaining 20% were selected for the validation.

| Extension of PNPP
The multivariate stepwise method was used to simulate PNPP. This method has been successfully used to estimate the sensitivity of NPP to climate factors (Luo et al., 2018). The annual mean ANPP, NDVI, temperature (T), precipitation (P), annual total solar radiation (SOL), annual potential evapotranspiration (PET), and DEM data were extracted from the undisturbed grids and normalized. Using the SPSS software (Statistical Product and Service Solutions, version 19.0), we defined the undisturbed ANPP as the dependent variable and other factors as independent variables, and then added these factors individually and deleted the unimportant variables to obtain the regression model. The PNPPs were then upscaled to the entire area based on this regression model.

| Calculation of NPP disturbed by human activities on NPP (HNPP)
HNPP was calculated using the following equation: where an HNPP value >0 indicates that human activities have a positive effect on the grassland NPP, while an HNPP value <0 indicates that human activities have a negative effect. For each pixel, the interannual change rate for HNPP was calculated using unitary linear regression analysis (Chen et al., 2019) as follows: where n represents the years (the time series was from 2001 to 2017, n = 17) and HNPP i represents the HNPP in year i. When HNPP > 0, θ slope > 0 represents a positive influence with an increasing trend (++), and + − indicates a positive influence with a decreasing trend; when HNPP < 0, θ slope > 0 represents a decreasing trend with a negative effect (−+). θ slope < 0 represents an increasing trend with a negative effect (−−). The F test was used for to test for significance.

| The relative contribution index (RCI) of human activities
Using the HNPP and PNPP, the RCI (Wu et al., 2017) of human activities was calculated to quantitatively describe the impact of human activities on grassland productivity. The equation is as follows: A higher RCI value signifies a greater impact of human activities on grassland productivity. RCI > 0.5 indicates that human activities is a dominant factor affecting grassland productivity.

| Validation of the CASA model
The field-observed data from 2017 were compared with the simulated values from the CASA model to verify the accuracy (Figure 2). The correlation between the observed and simulated data indicated that the CASA model performed well in modeling the NPP in this region.

| Distribution of undisturbed grids
After extracting the grids of (CR y > 0.8) ∩ (R m = 1) and excluding the grids that did not meet the requirements in the field sampling survey, the remaining grids (7213 in total) were considered as undisturbed grids. Most of these grids were distributed in the Qilian Mountain National Nature Reserve, with a few in the desert area. Grids downstream of the three basins were rare and unevenly distributed (Figure 3).

| Temporal and spatial distribution of ANPP, PNPP, and HNPP
The spatial distribution pattern of the ANPP in this area decreased from southeast to northwest and from 2001 to 2017 (Figure 4). Overall, the ANPP was relatively high in the eastern and south-eastern areas and the central oases but was low in the central plain and desert area. The spatial distribution of the PNPP was similar to the ANPP ( Figure 5). The areas with positive HNPP were mostly distributed in the southeast and along the Qilian Mountains. Small negative HNPP (2.35%) values were primarily distributed along the Qilian Mountains, moderate negative values (69.81%) mainly extended along the Qilian Mountains to the central plain area, and severe negative values (24.84%) were mostly distributed in the middle and lower reaches of the three major basins and the southwest area ( Figure 6).
The mean ANPP was 225.16 gC m −2 a −1 (Figure 7 Figure 8). The annual mean HNPP was negative, and the negative influence gradually decreased during the study period with a value of 2.79 gC m −2 a −1 (p < 0.05).
A positive HNPP occupied 1.89% (p < 0.05) of the grassland, which was distributed in the Qilian Mountains F I G U R E 2 Relationship between the CASA simulated NPP and observed NPP. NPP, net primary production. and the central plains and oases. A decreasing trend with a negative HNPP occupied 47.55% of the regions shown, primarily in the Qilian Mountains area and the northwestern area. The areas with a negative HNPP that increased accounted for 0.27% of the total area, which was mostly distributed in the northeast. The trend and direction of the HNPP in 50.29% of the region were not significant (Figure 9). F I G U R E 3 Spatial distribution of grids that are undisturbed by humans.

| RCI of human activities
From 2001 to 2017, the mean RCI values were all greater than 0.5, indicating that human activities were a dominant factor in the grassland NPP changes. The RCI decreased (p < 0.01) from 0.708 to 0.522 during the study period (Figure 10a). The area dominated by human activities (RCI > 0.5) decreased at a rate of 0.9% a −1 (p < 0.01) (Figure 10b), indicating that the contribution ratio of human activities was decreasing.
F I G U R E 5 Spatial distribution of the annual mean PNPP. PNPP, potential net primary productivity.
F I G U R E 6 Spatial distribution of annual mean HNPP. HNPP, human activities on net primary production.
The area with RCI < 0.5 accounted for 33.2% of the grassland (Figure 11), which was predominantly distributed in the Qilian Mountains where human activities were rare, and the ANPP change was primarily attributed to climate change. The area with 0.5 < RCI < 0.8 accounted for 36.87%, which was distributed in the central plains and desert areas that are densely populated. The area with RCI > 0.8 accounted for 29.93% and was mostly distributed in the desert. These areas were dominated by human activities (RCI > 0.5), with a total area that was greater than 66.8% (Figure 11).

| Validation of the modeled NPP
In this study, the mean ANPP for the grassland in the Hexi Corridor was 225.16 gC m −2 a −1 . We compared our results with other reports in arid and semi-arid grasslands. For example, the annual mean NPP was 260.44 gC m −2 a −1 in Inner Mongolia from 2003 to 2008 (J. Li et al., 2013b). In the Heihe River basin, the grassland NPP was 150-200 gC m −2 a −1 from 2001 to 2010 (Zhou et al., 2013), and in northwest China, the grassland NPP was 100-400 gC m −2 a −1 (Zhang et al., 2014). These results were comparable to our results, indicating that the estimation in this study is reliable. In addition, it has been reported that the NPP has increased significantly in the Hexi Corridor (Han et al., 2014), and NDVI in Northwest China has increased over the past two decades (Tang et al., 2017). These findings were consistent with our results that the ANPP was increasing (Figure 7).

| Undisturbed grids
The distribution of undisturbed grids is uneven (Figure 3). These grids were predominantly distributed in the middle and south-eastern areas, where water resources are abundant, with significant grassland growth. These areas were slightly affected by human activities, with negligible impacts on NPP. Vegetation in this area strongly depends on the climate and is sensitive to climate change (Grumbine, 2014). In contrast, there were few undisturbed grids in the desert areas. Although human activities may have affected these areas, the vegetation cover was especially low, and many areas were characterized by barren land due to the extreme drought conditions . Therefore, it is reasonable that the changes in the grassland NPP were not significant in these areas. From the spatial distribution of the undisturbed grids, it F I G U R E 7 The annual mean changes in ANPP and PNPP. ANPP, actual net primary production; PNPP, potential net primary productivity.
F I G U R E 8 The annual mean changes in HNPP. HNPP, human activities on net primary production. WANG and DOU | 217 can be concluded that our methods to extract the grids that were not affected by human activities were reliable.

| Temporal and spatial variation in HNPP
Positive HNPP values were only found in small areas (3%) (Figure 6), and those with significant increasing trends only occurred in 1.89% of the area, such as the Qilian Mountains and the oasis edge. In these areas, ecological measures, such as the Grain for Green Program and artificial planting, were implemented in the 2000s. These results suggested that water resources have been conserved, further leading to a continuously positive effect on vegetation owing to the more abundant soil water conditions (Buus-Hinkler et al., 2006;Ide & Oguma, 2013;Tang et al., 2017).
In most areas, the human influence was negative, which was attributed to human activities putting pressure on the grasslands O'Mara, 2012). The National Nature Reserve is located in the Qilian Mountains, and there are still some human activities, such as mining, F I G U R E 9 Interannual trend and direction of HNPP ("−+" represents a negative HNPP with decreasing trend, with 28.44% of the total area passing the significant test at 0.01 level; "−−" represents a negative HNPP with an increasing trend, and 0.04% of the total area passed the significant test at 0.01 level; "++" represents a positive HNPP with an increasing trend, and 1% of the total area passed the significant test at 0.01 level; "+−" represents a positive HNPP with a decreasing trend. The gray areas indicate the areas without significant changes). HNPP, human activities on net primary productivity.
F I G U R E 10 Changes in the mean RCI (a) and proportion of RCI > 0.5 (b). RCI, relative contribution index. herdsmen, and artificial fires, which were the causes of the negative impacts. In the areas with abundant water resources, human activities were intensive, which resulted in a negative influence (Olsson et al., 2005). In addition, the number of livestock in the Hexi Corridor increased from 2001 to 2010 (Han et al., 2014), and overgrazing led to grassland degradation (Schönbach et al., 2011). The middle and lower reaches in the basin are desert areas with a lower population density. Two reasons may explain the negative impacts. First, the reservoirs constructed in the middle reaches intercept the surface water otherwise flowing toward the lower reaches; Second, excessive amounts of groundwater were extracted to irrigate cultivated land, which led to a decline in the water table and serious grassland degradation (Mwendera et al., 1997;Piao et al., 2004). Most of the negative HNPP in the study area showed a decreasing trend because human activities have decreased due to the implementation of ecological projects, including the Grain for Green Project, pasture grazing rotation, ecological water transport, and ecological migration (F. Li et al., 2013a).

| RCI analysis
The areas with lower RCI values (RCI < 0.5) were primarily distributed in the southern Qilian Mountains, where human activities were rare due to the limitation of climate, traffic, and topography. Generally, vegetation in mountainous areas is more sensitive to climate change than other ecosystem (Anderson & Goulden, 2011). The RCI was high in the Gobi Desert, where the ANPP was low and vulnerable to human activities, and slight human activities could lead to a high RCI. The areas with RCI > 0.5 showed the most obvious decreasing trends in the central plains area, where government has implemented a centralized grazing plan. These results suggested that these projects effectively decreased the negative effects of human activities on the NPP (F. Li, Zhu, et al., 2013a;Zhu & Li, 2014).

| CONCLUSIONS
In this study, the anthropogenic impacts on the grassland NPP in the Hexi Corridor from 2001 to 2017 were quantitatively evaluated. The highlight of this study is that we proposed a novel PNPP estimation method based on undisturbed grids. The results showed that most of those grids were distributed in the Qilian Mountain National Nature Reserve, with a few in the desert area, and the calculation of PNPP based on ANPP was similar to the natural conditions and improved the reliability of the HNPP estimation. During the study period, the mean ANPP, PNPP and HNPP were 225.16 gC m −2 a −1 , 449.77 gC m −2 a −1 respectively, and the overall annual mean HNPP was negative with −236.25 gC m −2 a −1 . After the implementation of environmental protection projects, the negative impact of human activities decreased significantly, and the RCI of human activities also decreased. Based on the identification of grids without human activities, we assessed the effects of human activities on the grassland NPP. Our results suggested that human activities was a more significant factor in the determination of grassland NPP than climate, and the implementation of ecological protection measures benefits ecological restoration in the Hexi Corridor. These findings highlight that human activities must be considered regarding ecosystem changes in arid and semi-arid regions.

ACKNOWLEDGMENTS
We would like to thank Chuanhua Li for his guidance and suggestions on research methods, and Xuemei Yang for providing the field data. This study was financially supported by the National Natural Science Foundation of China (41761083) and Soft Science Research Project of Gansu Province (21CX6ZA020).

DATA AVAILABILITY STATEMENT
A peer review of empirical data will be conducted to confirm that the data reproduce the analytic results reported in the article.

ETHICS STATEMENT
None declared.