Changes in Net Primary Productivity and Factor Detection in China’s Yellow River Basin from 2000 to 2019

Abstract

Net primary productivity (NPP) is a main contributor to ecosystem carbon pools. It is crucial to monitor the spatial and temporal dynamics of NPP, as well as to assess the impacts of climate change and human activities to cope with global change. The dynamic of the NPP in China’s Yellow River Basin (YRB) from 2000 to 2019 and its influencing factors were analyzed by using trend and persistence tests and the GeoDetector method. The results show that the NPP had strong spatial heterogeneity, with a low NPP in the west and north, and a high NPP in the east and south. From 2000 to 2019, the NPP showed a statistically significant increase (at a mean of 5.5 g C m⁻² yr⁻¹, for a cumulative increase of 94.5 Tg C). A Hurst analysis showed that for the NPP in 76.3% of the YRB, the time series was anti-persistent. The spatial heterogeneity of the NPP in the YRB was mainly explained by precipitation and relative humidity (q value ranged from 0.24 to 0.44). However, the strength of the precipitation explained the decreased variation over time (q value decreased from 0.40 in 2000 to 0.26 in 2019). Interactions between the climate factors and human activities affected the NPP more strongly than individual factors. The results emphasize the importance of strengthening future research on the interaction between climate change and human activities. The results reveal the risk and optimal ranges of the driving factors and provide a quantification of the impacts of those factors regarding NPP. These findings can provide a scientific basis for vegetation restoration in the YRB.

1. Introduction

Vegetation is an important foundation of the terrestrial ecosystem and links energy and material transfer. Net primary productivity (NPP) is defined as the net photosynthetic carbon assimilation after subtracting the consumption of carbon through autotrophic respiration. NPP is an important indicator for the material cycle (especially the carbon and water cycle) [1,2], energy flows [3], and ecosystem quality [4]. NPP is the main source of carbon sinks for terrestrial ecosystems, and promoting increased NPP is a key aspect of mitigating climate change [5]. However, due to the simultaneous influence of climate change and human activities, ecosystems in dry areas are at increasing risk of degradation, with increased effects such as decreased vegetation cover, carbon and nutrients loss, and desertification [6,7]. These processes can change an ecosystem from a carbon sink to a carbon source [8].

Increased precipitation generally has a positive impact on NPP because plant growth in many areas is most strongly restricted by long-term water availability [9]. NPP shows an especially strong correlation with the precipitation in drylands that is widely distributed around the world [10]. Previous research confirmed that rainfall variability is a key constraint on the NPP in China [11]. In contrast, the impact of temperature on plant growth is less certain. On the one hand, increasing temperatures increase evaporation and can cause drought that is severe enough to decrease NPP [8,12]. On the other hand, increasing temperatures can prolong the growing period, thus promoting increased NPP [13]. Previous studies showed that vegetation is mainly affected by precipitation in arid and semi-arid areas, but it is mainly affected by the temperature in alpine regions [7,14]. However, other researchers have reported that the change in vegetation productivity in both semi-arid and alpine regions was dominated by temperature [15]. Despite these contradictory findings, these studies show that the vegetation in arid, semi-arid, and alpine regions is increasingly vulnerable to the impact of climate change.

Similar to climate change, human activities also have an important impact on vegetation. These impacts result from population growth, agricultural and urban expansion, grazing of livestock, logging, and ecological conservation [16,17,18]. These processes often lead to fundamental changes in vegetation types. Widespread vegetation degradation has been observed around the world, especially in drylands, and land use change is the main reason for vegetation changes [6,19,20,21]. The increasing atmospheric carbon dioxide concentration caused by human activities also directly and indirectly affects plant growth through the resulting climate change [22]. At the same time, targeted vegetation restoration measures are being carried out around the world to cope with increasingly severe environmental changes. For example, vegetation restoration, which is a major activity in China, is being implemented through large-scale afforestation projects and strict ecological protection measures. Such measures have been responsible for 25% of vegetation restoration while located in only 6.6% of the global vegetation area [23,24]. Ecological restoration increases vegetation biomass and an ecosystem’s carbon sink [25], and directly or indirectly affects soil carbon storage [26].

The impact of environmental change on vegetation is not always unidirectional. Vegetation can change the surface temperature and energy balance by changing the surface reflectance and evapotranspiration [3,27,28,29]. Vegetation can also change the environmental effect of precipitation by changing interception and evapotranspiration, thereby changing the water availability (e.g., drought and changes in runoff and sediment transport) [1,30,31]. More importantly, the feedback between vegetation dynamics and climate change can strengthen the spatial heterogeneity of vegetation [32]. The resulting NPP changes directly affect the carbon sequestration of ecosystems and global environmental change. Therefore, it is urgently necessary to monitor the spatial and temporal changes of NPP.

China has many large river basins whose ecosystems are responding to climate change, among these is the Yellow River Basin (YRB). The YRB covers many ecologically fragile areas with serious vegetation degradation and soil erosion, such as the Mu Us Sandy Land, the Hobq Desert, the Ulan Buh Desert, the Qinghai–Tibet Plateau, and the Loess Plateau. The occurrence of degradation before ecological protection was implemented in these areas has made the Yellow River have the greatest sediment load among the world’s rivers [33]. Fortunately, a series of ecological projects in the YRB have been implemented since the 1950s. These projects have led to obvious vegetation restoration and decreased sediment load, especially after the widespread implementation of “Grain for Green” in 2002 for afforestation on sloping farmland and wasteland, which are prone to soil erosion [33,34,35]. Previous studies have reported the characteristics of vegetation dynamics and the natural and artificial driving factors [7,11,36]. However, our understanding the dynamic of NPP and its driving factors at the basin scale remains incomplete, especially the risk and optimal ranges and the interaction between natural and human factors.

To enhance our knowledge and to fill in the missing information, we designed this study to analyze the spatial and temporal changes of NPP and its mechanisms in the YRB during 2000–2019. We hypothesized that (1) human activities have a more substantial and direct impact than climatic drivers on NPP changes, and (2) there is an interaction between human activities and climatic drivers. We asked the following questions: (1) Does the NPP in the YRB show spatiotemporal heterogeneity? (2) Do human activities have a stronger impact than climatic drivers on NPP in this region? (3) Is there an optimal range of driving factors that affect NPP? The goal was to analyze the relationship between NPP heterogeneity and environmental factors, and to determine the potential risk and the optimal ranges of environmental factors for NPP. The results can provide a more comprehensive understanding of the vegetation carbon sink in this key region of China and serve as a reference for the formulation of more effective regional ecological restoration strategies.

2. Materials and Methods

2.1. Study Area

The total length of the Yellow River is 5464 km and the YRB covers 9 provinces (including autonomous regions) in China, with a total drainage area of 795,000 km². The YRB is located between 32.16°N and 41.86°N and between 95.88°E and 119.07°E (Figure 1). The YRB is dominated by arid and semi-arid areas, and it includes semi-humid areas in the southeast. The climate types are mainly a plateau climate (mainly located west of Lanzhou) and a temperate continental climate. The precipitation is high in the south of the YRB (Figure S1), while the temperature is high in the east (Figure S2). The average total annual precipitation is 476 mm, and the annual average temperature is 6.4 °C. The landforms of the YRB mainly include plateaus, mountains, loess, and alluvial plains, with the highest elevations in the west and lowest elevations in the east. The vegetation types present are mainly alpine meadow, temperate grassland, warm temperate deciduous broad-leaved forest, and crops (Figure S3) [37].

2.2. Data Collection

We obtained NPP data (2000–2019) with a spatial resolution of 500 m from the website of NASA’s Terra MODIS MOD17A3 products (https://ladsweb.modaps.eosdis.nasa.gov/, accessed on 21 April 2022). Annual average temperature and total precipitation (both with a spatial resolution of 1 km), wind speed, solar radiation, and relative humidity (with a spatial resolution of 0.1°) were obtained from the National Tibetan Plateau Data Center (https://data.tpdc.ac.cn/, accessed on 19 April 2022). In addition, they were verified using the independent meteorological observation point data that were obtained by the producer. Land-use types (with a spatial resolution of 1 km), elevation (with a spatial resolution of 30 m), and population and gross domestic product (GDP) density (population and GDP per unit area (1 km²), both with a spatial resolution of 1 km) were provided by the Data Center for Resource and Environment Science (http://www.resdc.cn/, accessed on 16 July 2022). Except for the NPP data from 2000 to 2019, the other data were from 2000, 2005, 2010, 2015, and 2019. All data were reprojected to the same projection coordinate system and resampled by bilinear interpolation to a uniform resolution of 1 km using ArcGIS 10.2 (www.esri.com, accessed on 25 April 2022). Scatterplots, linear fitting, and visualization were performed using Origin 2018 (https://www.originlab.com/, accessed on 25 April 2022).

2.3. Methods

2.3.1. Change Trend Detection

We used a linear regression slope to analyze the changes in the NPP from 2000 to 2019 in each pixel with Equation (1). We combined the Theil–Sen median and the Mann–Kendall methods to detect the NPP trends by using Equations (2)–(5) [38,39,40]:

$$slope=\frac{n {\displaystyle \sum _{i=1}^n\left(i\times NP{P}_i\right)-{\displaystyle \sum _{i=1}^{n}i {\displaystyle \sum _{i=1}^{n}NP{P}_i}}}}{n {\displaystyle \sum _{i=1}^n{i}^2-{\left({\displaystyle \sum _{i=1}^{n}i}\right)}^2}}$$

(1)

$$\mathsf{\beta }=\mathrm{median}\left(\frac{NP{P}_j-NP{P}_i}{j-i}\right)$$

(2)

$$\mathrm{sgn}\left(NP{P}_j-NP{P}_i\right)=\left\{\begin{array}{c}1 \mathrm{for} NP{P}_j-NP{P}_i>0\\ 0 \mathrm{for} NP{P}_j-NP{P}_i=0\\ -1 \mathrm{for} NP{P}_j-NP{P}_i<0\end{array}\right.$$

(3)

$$S={\displaystyle \sum _{i=1}^{n-1}{\displaystyle \sum _{j=i+1}^n\mathrm{sgn}\left(NP{P}_j-NP{P}_i\right)}}$$

(4)

$$Z=\left\{\begin{array}{c}\frac{S-1}{\sqrt{\mathrm{var}\left(S\right)}} \mathrm{for} S>0\\ 0 \mathrm{for} S=0\\ \frac{S+1}{\sqrt{\mathrm{var}\left(S\right)}} \mathrm{for} S<0\end{array}\right. ; \mathrm{var}\left(S\right)=\frac{n\left(n-1\right)\left(2n+5\right)}{18}$$

(5)

where n is the length of the study period; NPP_i and NPP_j represent the NPP time series i and j (1 ≤ I < j ≤ n); β represents the Theil–Sen median, which shows an increasing NPP for a positive β, and a decreasing NPP for a negative β; var(S) represents the variance of the test statistic S; and the Z statistic is defined for trend significance testing. At the given significance level (α = 0.05), |Z| > 1.96 represents a statistically significant change trend.

2.3.2. Hurst Analysis

The Hurst exponent (H) is an effective tool for assessing the persistence of a time series (i.e., the likelihood that a high value will be followed by a high value and a low value by a low value) [41]. We used H to estimate future changes of NPP. To do so, we decomposed the time series of NPP_i values into non-overlapping subsequences NPP_t with length δ, and calculated the H value using Equations (6)–(10) [42,43]:

$$\overline{NP{P}_{\mathsf{\delta }}}=\frac{1}{\mathsf{\delta }}{\displaystyle \sum _{t=1}^{\mathsf{\delta }}NP{P}_t} ; \left( 1\le t\le \mathsf{\delta }\le n\right)$$

(6)

$$U_{t\mathsf{\delta }}={\displaystyle \sum _{t=1}^{\mathsf{\delta }}\left(NP{P}_t-\overline{NP{P}_{\mathsf{\delta }}}\right)}$$

(7)

$$R_{\mathsf{\delta }}=\max U_{t\mathsf{\delta }}-\min U_{t\mathsf{\delta }}$$

(8)

$$S_{\mathsf{\delta }}=\sqrt{\frac{1}{\mathsf{\delta }}{\displaystyle \sum _{t=1}^{\mathsf{\delta }}{\left(NP{P}_t-\overline{NP{P}_{\mathsf{\delta }}}\right)}^2}}$$

(9)

$$R_{\mathsf{\delta }}/S_{\mathsf{\delta }}={\left(c\mathsf{\delta }\right)}^H$$

(10)

where NPP_t is a time series and δ ≥ t; U_tδ represents cumulative deviation; R_δ and S_δ represent the range and standard deviation sequences, respectively; and c and H (0 < H < 1) are the fitting parameters. Furthermore, 0 < H < 0.5 indicates the anti-persistence of NPP (i.e., the opposite trend to the past, and the lower the value of H, the stronger the anti-persistence), H = 0.5 indicates a random series, and 0.5 < H < 1 indicates a persistence (the higher the value of H, the stronger the persistence).

2.3.3. Geographic Detector

We used GeoDetector (http://www.geodetector.cn/, accessed on 11 November 2022) to assess the spatial heterogeneity of NPP and its driving factors. We measured heterogeneity using the q value in Equation (11) [44,45]:

$$q=1-({\displaystyle \sum _{h=1}^L{N}_h{\mathsf{\sigma }}_h^2)}/ (N{\mathsf{\sigma }}^2) ; (0\le q\le 1)$$

(11)

where the q value indicates that the environmental factor explains 100 × q% of the NPP heterogeneity in this study (for example, when the q value is 0.5, the explanation rate is 50%, and the higher the value of q, the greater the heterogeneity of NPP or the stronger the influence of environmental factors on NPP); h represents the stratification of the response variable (NPP) or environmental factor (X); N_h and N represent the numbers of units in stratum h and the whole region, respectively; and σ_h² and σ² represent the corresponding variances of NPP.

After obtaining the q value for each environmental factor, we conducted factor–pair interaction detection—that is, we calculated the interaction between variables X_a and X_b by superimposing these variables to form a new distribution. The interaction effect between environmental factors was identified by the interactive q value. We applied the risk region detector to analyze the difference in mean value between the sub-regions by testing with the t-statistic and detecting the optimal range of factors for NPP in Equation (12):

$$t_{{\overline{Y}}_{h=i}-{\overline{Y}}_{h=j}}=\frac{{\overline{Y}}_{h=i}-{\overline{Y}}_{h=j}}{{\left[\mathrm{Var}\left({\overline{Y}}_{h=i}\right)/n_{h=i}+\mathrm{Var}\left({\overline{Y}}_{h=j}\right)/n_{h=j}\right]}^{\frac{1}{2}}}$$

(12)

where Y_h=i and Y_h=j are the mean NPP of sub-region i and j, respectively; n is the sample size; and Var is the variance.

To implement GeoDetector, we discretized the continuous environmental variables (elevation, precipitation, temperature, and population and GDP density) using the natural breaks method to ensure maximum differences between the categories [46]. To avoid overflow in GeoDetector [45], we generated 2024 samples based on 20 km × 20 km grids.

3. Results

3.1. Spatial and Temporal Changes of NPP

We used the arithmetic average of the NPP in the YRB during the 20-year study period to obtain the pattern of NPP (Figure 2a). The NPP in the YRB showed a strong spatial heterogeneity, with low values in the northern and western areas and high values in the southern and eastern areas. The areas with high NPP (>400 g C m⁻² yr⁻¹) were mainly located in the southern YRB (an area with mainly warm-temperate mountain forest), and these accounted for approximately 20% of the total area. The areas with low NPP (<100 g C m⁻² yr⁻¹) were mainly located in the western and northern YRB (e.g., alpine desert grassland and the Hobq Desert), which accounted for 5.6% of the total area.

From 2000 (247 g C m⁻² yr⁻¹) to 2019 (358 g C m⁻² yr⁻¹), the annual mean NPP increased statistical significantly (p < 0.001) with considerable fluctuation (Figure 2b). The mean NPP for the 20 years was 316 g C m⁻² yr⁻¹. The NPP increased by 5.5 g C m⁻² yr⁻¹ on average, with a mean annual growth rate of 2.2%. The NPP was largest in 2018 (379 g C m⁻² yr⁻¹), and lowest in 2001 (243 g C m⁻² yr⁻¹). During the 20-year study period, the NPP increased fastest in 2002, 2012, and 2018, with a growth rate of more than 12%, whereas the years with a decreased mean NPP accounted for 45%. The total NPP of the YRB rose from 195 Tg C yr⁻¹ in 2000 to 289 Tg C yr⁻¹ in 2019.

3.2. Trend and Persistence of NPP

We calculated the pattern of the slope (i.e., the average rate of change) of the NPP in the YRB using the NPP series from 2000 to 2019 (Figure 3a). From 2000 to 2019, the statistically significant (|Z| > 1.96) increased the NPP and was mainly distributed in the middle, accounting for 78.8% of the YRB. The areas with rapid NPP growth (>6 g C m⁻² yr⁻¹) were mainly located in the central and eastern YRB. The areas with statistically significantly decreased NPP (|Z| > 1.96) were mainly located in the built-up areas of cities (Figure 3a, in red), accounting for 0.3% of the basin. The areas with non-significant changes or no change were mainly distributed in western, eastern, and northern parts of the YRB, accounting for 20.9%. Hurst analysis showed that the NPP of the basin mainly had an anti-persistent time series in 76.3% of the basin (Figure 3b). The areas with a persistent NPP time series were mainly located in the middle of the basin, accounting for 20.9%.

3.3. Factors Influencing NPP Heterogeneity

3.3.1. Factor Detection

The factor detection analysis indicated that climatic factors, especially precipitation and relative humidity, were the key environmental factors that affected the spatial heterogeneity of NPP (

Table 1. Factors for changes in the NPP in YRB based on GeoDetector q values.

Factors	2000	2005	2010	2015	2019
Land use	0.12 **	0.14 **	0.13 **	0.16 **	0.13 **
GDP density	0.02	0.01	0.01	0.02 *	0.02 *
Population density	0.04 **	0.04 **	0.02	0.09 **	0.04 **
Precipitation	0.40 **	0.36 **	0.36 **	0.30 **	0.26 **
Temperature	0.19 **	0.17 **	0.12 **	0.15 **	0.12 **
Wind speed	0.22 **	0.28 **	0.27 **	0.19 **	0.16 **
Relative humidity	0.37 **	0.38 **	0.39 **	0.44 **	0.24 **
Solar radiation	0.17 **	0.24 **	0.24 **	0.31 **	0.25 **
Elevation	0.15 **	0.14 **	0.10 **	0.12 **	0.11 **

Statistical significance: *, p < 0.05; **, p < 0.01.

). However, the ability of precipitation to explain NPP decreased over time, from 40% in 2000 to 26% in 2019 (p < 0.01). The ability of relative humidity to explain NPP fluctuated, with q values ranging from 0.24 to 0.44 (p < 0.01). The temperature, wind speed, and solar radiation were also statistically significant (p < 0.01) and important factors that affected the NPP, and they explained 12 to 31% of the spatial heterogeneity. In contrast, anthropogenic factors and elevation had a lower ability to explain the NPP (with q values ranging from 0.01 to 0.16) than climatic factors.

3.3.2. Interactions among Factors

Interaction detection among the factors showed that there were interactions between the environmental factors (

Table 2. Average interaction of factors in the YRB based on GeoDetector q values.

Factors	Land Use	GDP Density	Population Density	Precipitation	Temperature	Wind Speed	Relative Humidity	Solar Radiation	Elevation
Land use	0.14
GDP density	0.15	0.02
Population density	0.18	0.06	0.05
Precipitation	0.41	0.36	0.36	0.33
Temperature	0.26	0.16	0.18	0.48	0.15
Wind speed	0.32	0.26	0.28	0.40	0.37	0.22
Relative humidity	0.42	0.39	0.39	0.42	0.47	0.43	0.37
Solar radiation	0.31	0.27	0.27	0.47	0.36	0.38	0.46	0.24
Elevation	0.25	0.14	0.16	0.49	0.20	0.36	0.47	0.37	0.13

). In most cases, the interaction of pairs of driving factors was greater than the strength of the individual factors, indicating that both factors acted simultaneously on the NPP (i.e., bi-factor enhancement). For example, precipitation showed a strong interaction with temperature, elevation, and solar radiation, with interaction q values of 0.48, 0.49, and 0.47, respectively. Although the ability of anthropogenic factors to explain NPP is relatively low, there is a strong interaction between human activities and climatic factors. For example, land use, precipitation, and relative humidity alone had individual effects of 0.14, 0.33, and 0.37 on average, respectively. However, the interaction effects between land use and precipitation, and between land use and relative humidity increased to 0.41 and 0.42 on average, respectively. This means that the combination of land use and precipitation, and of land use and relative humidity explained 41 and 42% of the variation of NPP, respectively. This represents about 3, 1.2, and 1.1 times the ability of land use, precipitation, and relative humidity, respectively, to separately explain NPP.

3.3.3. Risk Detection and Optimal Ranges of Factors

The higher the NPP, the more suitable the area is for vegetation growth at the corresponding range of factor values (Figure 4). Among different land use types, NPP was the highest in forest. NPP increased first and then decreased with increasing GDP density and population density. The optimal range of GDP and population density was 8238.0–22,179.3 RMB km⁻² and 1182.8–3548.2 persons km⁻², respectively. Overall, the NPP increased with increasing precipitation, temperature, and relative humidity, which had optimal ranges of 676.2–973.0 mm, 11.7–15.9 °C, and 64.0–75.3%, respectively. The NPP decreased with increasing wind speed, solar radiation, and elevation, and the optimal ranges were 1.2–1.8 m s⁻¹, 139.0–159.5 W m^–2, and −7–697 m, respectively. In summary, the response of the NPP in the YRB to environmental factors was diverse and non-linear.

4. Discussion

4.1. Spatial and Temporal Dynamics of the NPP and Potential Impacts

The NPP in the YRB showed obvious spatial differences, with a gradient from north to south, which is consistent with previous instances of monitoring vegetation coverage [34,47]. The spatial change of the NPP was consistent with the pattern of vegetation types (Figure S3). Specifically, from north to south, temperature and precipitation gradually increase (Figures S1 and S2), resulting in a change in vegetation types. That is, vegetation types gradually change from desert and desert grassland to warm-temperate deciduous broad-leaved forest and crops. From west to east, the terrain (and particularly elevation) causes changes in the hydrothermal conditions, resulting in changes in vegetation from alpine desert grassland and alpine meadows to forest [47,48]. In terms of human activities, the cropland is mainly in the southeastern YRB (with a semi-humid climate) (Figure S3). The NPP in the cropland was higher than that in the desert and grassland areas, but lower than that in the forest (with a semi-humid climate) (Figure 4). The patterns and values of the NPP under different levels of vegetation cover are consistent with previous research. For example, the NPP was generally less than 100 g C m⁻² in arid deserts and desert grassland [49].

In 78.8% of the YRB, NPP showed a statistically significant increase over time, and in 20.9% it showed a non-significant change, which is in accordance with previous works on vegetation [48,50]. The NPP estimation in the YRB by a Boreal Ecosystem Productivity Simulator model showed that the NPP increased by an average of 2.35 g C m⁻² yr⁻¹ during 1981–2020 [50], versus 5.5 g C m⁻² yr⁻¹ during 2000–2019 in the present study. Our results indicate that the vegetation restoration in the YRB during 2000–2019 accelerated, reaching several times the previous levels. Precipitation increased (by about 3.8 mm yr⁻¹ on average) in the past 20 years (Figure S1b), and the soil moisture increased (0.05 to 0.07% in volume) [35], which may have led to rapid vegetation restoration because long-term water availability is an important factor affecting vegetation growth [9]. The NPP increased rapidly (>6 g C m⁻² yr⁻¹) and was mainly located in the mountainous and hilly areas in the central and eastern parts of the study area, coinciding with the area where desertified land was green and cropland was transformed into grassland and forest (Figure S4). Therefore, vegetation restoration in the YRB may be mainly related to land use change. Other studies have confirmed that afforestation has been the main restoration technique used in northern China [2,24]. Specifically, the speed of vegetation restoration in the YRB after afforestation on sloping cropland and wasteland greatly exceeded those in the grassland in northern China (1.7 g C m⁻² yr⁻¹) [51] and in China as a whole (3.13 g C m⁻² yr⁻¹) [7].

A direct benefit of vegetation restoration is increased carbon storage. According to our estimates, the total NPP increased by 94.5 Tg C from 2000 to 2019, for a total increase of 48.5% over 20 years. Although we did not estimate the soil carbon storage in the present study, previous research [26] strongly suggests that vegetation restoration will increase soil carbon storage. Vegetation restoration would therefore help China achieve the goal of carbon neutrality. Restoration would also mitigate climate change. Furthermore, an increased plant carbon pool would directly decrease the atmospheric CO₂ concentration. In addition, the cooling effect that results from increased evapotranspiration would reduce the local temperature [29,52]. The cooling caused by the restoration of forest vegetation would be especially obvious [27]. In addition, vegetation restoration is the key reason for decreased water and soil loss in the YRB [33].

Although vegetation restoration is of great significance, it still draws attention to the adverse impact of such change on the ecosystems in the basin. An estimate based on the water demand for vegetation restoration and human activities showed that the maximum sustainable NPP in the Loess Plateau (one of the most important geomorphic units in the YRB) was 400 ± 5 g C m⁻² yr⁻¹ [53]. A water shortage will occur when the NPP is greater than this threshold [53]. Zhang et al. [54] found that the vegetation coverage in the central and eastern Loess Plateau has exceeded the climatic water threshold. This problem occurs because increased plant water demand beyond the level supplied by the soil results from the increased evapotranspiration caused by vegetation restoration [52,54]. The NPP in the YRB in our study approached this threshold (reaching 378.9 g C m⁻² yr⁻¹ in 2018), and it may have already exceeded the threshold created by water availability in the eastern and southern regions. These results indicate that, against the background of climate change, excessive afforestation in the YRB could lead to more serious drought. Therefore, more sustainable restoration schemes, such as restoration based on shrubs and grasses with high water-use efficiency, will be needed for future ecological restoration.

4.2. Impact of Climate and Human Activities on NPP

Our analysis revealed that the spatial heterogeneity of the NPP in the YRB was mainly related to climate factors that changed the hydrothermal conditions. Precipitation and relative humidity were especially critical factors, and they explain the largest part of the NPP variation. Our results are different from those of Shi et al. [55], who found that the NPP in China was mainly affected by land use. These results emphasized the differences in the NPP patterns between large-scale regions and sub-regions. The YRB is dominated by an arid or semi-arid climate, which gradually transitions to a semi-humid climate in the southeastern and western parts of the basin; as such, plant growth heavily depends on the availability of water [7,11,47]. In the present study, we found that decreasing precipitation explains NPP variation since 2000, and this differs from a previous report suggesting that the ability of climate factors to explain this variation will increase against the background of climate change [7]. A possible reason for this difference is that the water-use efficiency of vegetation increases with increasing atmospheric CO₂ concentration and the promotion of effective human management, resulting in a decrease in the water demand of plants [56,57,58,59]. This increase in water-use efficiency is also conducive to increasing the adaptability of plants to regional climate change. Temperature affects plant growth mainly by affecting the plant’s phenology and physiological processes. Climate warming prolongs the growing season, especially in spring and autumn, and will promote early seed germination and subsequent plant photosynthesis. In contrast, summer warming may exacerbate drought and inhibit plant growth [13,59]. In alpine regions, warming may have a positive impact on plant growth [7,60]. Unfortunately, our results do not support previously reported results showing that temperature was the main factor affecting vegetation growth [35]. In contrast, the NPP decreased with increasing wind speed, solar radiation, and elevation (Figure 4). Increased solar radiation can cause reduced protection and stomatal closure [61,62]. In addition, increased solar radiation and wind speed can exacerbate drought and cause smaller plants [63]. Though elevation is a static variable, it has a greater impact than the dynamic variables of anthropogenic factors, such as GDP and population density. The large difference in elevation in the YRB (ranging from −7 to 6065 m) causes a redistribution of hydrothermal conditions, especially changes in temperature, leading to changes in vegetation types and plant growth [55].

Human activities have both negative and positive impacts on vegetation growth. For example, economic development, urban expansion, and increased population can threaten the health of vegetation because raw materials and land resources are needed to sustain development; conversely, economic development can also generate income that can be used to support ecological construction [64]. It is estimated that the optimal population density for vegetation restoration is 19–42 people km⁻² [64]. The average population density in the YRB increased from 142 people km⁻² in 2000 to 164 people km⁻² in 2019, which indicates that vegetation restoration may be threatened by population growth. The optimal range of population density in this study differs strongly from the aforementioned range, partly due to the different research subject and the improvement of the NPP through artificial management measures (e.g., artificial green spaces and intensive farming). However, excessive population growth undoubtedly has a negative impact on NPP (Figure 4) [55,64]. Previous studies have reported on the impacts of land use change on ecological restoration [21], especially in China, where the conversion of farmland to forest or grassland has accelerated vegetation restoration [47,65]. Although ecological restoration projects have accelerated vegetation restoration in the YRB, highlighting the strong impacts of human activities on NPP, our factor detection showed that the ability of individual human factors to explain the spatial pattern of NPP was low. This may be because land use change and population aggregation mainly occur in agricultural land areas and cities. For natural vegetation far from human settlements (such as at a high elevation and in mountain forests), human activities have less impact than climate factors on the spatial heterogeneity of NPP [7,66]. In addition, we found that the interactions between pairs of factors increased the ability of the pairs of climatic and human factors to explain NPP heterogeneity, which was consistent with previous vegetation monitoring results in the Loess Plateau [34,64]. The results therefore emphasize the necessity of determining the synergistic effects of climate change and human activities on NPP rather than focusing on individual factors in isolation.

4.3. Limitations

While we analyzed the spatio-temporal changes and influence factors of the NPP in the YRB, this study still has some limitations. First, the spatial resolution of the MODIS products is low, and coarse models may produce uncertain errors in areas with high heterogeneity, such as complex vegetation cover changes and mixed pixel regions [67]. Therefore, while considering efficiency, remote sensing images and models with higher spatio-temporal resolution should be used in future research on vegetation monitoring. Second, it is difficult to obtain long-term ground flux measurements to verify the reliability of the NPP products in the YRB. This is also a challenge faced by many remote sensing products, including MODIS NPP products [68]. Therefore, establishing sufficient sites for ground measurement to verify or evaluate remote sensing models will make it more convincing in future research. Third, we investigated the influence of nine environmental factors (including three anthropogenic factors) on the NPP in this study. This was mainly due to the limited data availability. In fact, there are far more factors that affect NPP, especially anthropogenic factors and soil properties. Therefore, research on human activities (e.g., grazing, deforestation, fires, and waste emissions) and soil properties (e.g., soil organic matter and nutrient availability) that affect NPP should be strengthened in the future.

5. Conclusions

In this study, we analyzed the spatial and temporal dynamic of NPP and the impact of climate and human activities on the NPP in the YRB. The results answer the questions we raised and partially support our hypothesis. Specifically, NPP showed strong heterogeneity, which increased from the northwest to the southeast in the YRB. The NPP time series during the 20-year study period increased rapidly overall, indicating substantial vegetation restoration in the YRB, which will promote progress toward achieving regional carbon neutrality. Climate and land use change (e.g., afforestation) have jointly promoted this recovery. Precipitation and relative humidity were the key factors explaining the heterogeneity of the NPP in the YRB, but the proportion of precipitation explaining the variation has been decreasing. The interactions between pairs of factors increased the ability to explain the variation. The results revealed the risk and the optimal ranges of the driving factors, as well as provide quantification of the impacts of those factors on the NPP. However, the response of the NPP to environmental factors was diverse and non-linear. The results emphasize the influence of climate factors and their interaction with human activities, as well as the importance of carefully implementing ecological protection projects to ensure that they are sustainable. To support such projects, it will be necessary to strengthen future research on the interactions between climate change and human activities.