Analysis of Dynamic Changes in Vegetation Net Primary Productivity and Its Driving Factors in the Two Regions North and South of the Hu Huanyong Line in China

Abstract

Human activities and global environmental changes have transformed terrestrial ecosystems, notably increasing vegetation greenness in China. However, this greening is less effective across the Hu Huanyong Line (Hu Line). This study analyzes dynamic changes and driving factors of nine vegetation net primary productivities (NPPs) in regions divided by the Hu Line using remote sensing data, trend analysis, and the Geodetector model. Findings reveal that from 2001 to 2022, 38.22% of regional vegetation NPP in China increased, especially in the Loess Plateau, Sichuan Basin, and Northeast Plains, while 2.39% decreased, primarily in the southeastern region and southern Tibet. Grasslands contributed 39.71% to NPP north of the Hu Line, and cultivated vegetation contributed 50.58% south. The driving explanatory power of factors on vegetation NPP on the north side of the Hu Line is generally greater than that on the south side. Natural factors primarily drive NPP changes, with human activities having less impact. Combined factors, particularly climate and elevation, significantly enhance the driving explanatory power (q, 0–1). The joint effects of elevation and precipitation on grassland NPP dynamics (q = 0.602) are notable. GDP’s influence on broadleaf forests north of the Hu Line (q = 0.404) is significant. Grasslands respond strongly to land use changes and population density, with a combined effect of q = 0.535. Shrubs, alpine vegetation, and meadows show minimal response to individual factors (q < 0.2). These findings offer insights for devising ecological protection measures tailored to local conditions.

1. Introduction

Vegetation net primary productivity (NPP) is a crucial indicator of ecosystem function, directly related to critical scientific issues of global change such as the carbon cycle, water cycle, and food security [1]. It reflects the ecosystem’s energy storage, the productive capacity, and the combined effects of climate change and human activities on terrestrial vegetation. It offers significant reference value for ecological conservation and sustainable development [2,3]. At present, domestic and foreign scholars have achieved fruitful results in research in the fields of NPP measurement methods [4,5,6], spatiotemporal change patterns [7,8,9], and driving factors [10,11,12]. In recent years, with the rapid development of remote sensing technology, vegetation NPP estimation methods based on remote sensing observations have been effectively used for monitoring annual fluctuations and long-term trends in NPP [8,13]. Particularly, the MODIS dataset from the TERRA satellite, based on the remote sensing BIOME-BGC model [14], currently plays a significant role in global studies on vegetation growth status, spatiotemporal change patterns, and responses to climatic factor changes [15,16].

Research shows that China’s terrestrial vegetation has become significantly greener in the past 20 years [17], and regional ecosystem structures and functions have undergone changes that cannot be ignored. Many scholars have studied the spatiotemporal changes and driving factors of vegetation NPP in various regions of China and found that China’s vegetation NPP has increased significantly in recent years and is affected by climate and human activity factors to varying degrees [18,19,20]. However, it is generally difficult to cross the Hu Line when greening vegetation in China [21]. As an iconic geographical boundary, the Hu Line serves as a dividing line for population distribution density in China and as a crucial natural ecological boundary. The Hu Line divides China into two drastically different ecosystems: the northwest and the southeast [22]. In different ecosystems, the spatial and temporal changes in vegetation productivity and their complex response mechanisms to human activities, the natural environment, and other factors are still hot issues in research. Therefore, in this context, this study chose to divide China into two regions by the Hu Line and applied the geographical detector model to analyze the main driving factors and geographical differences of dynamic changes in vegetation productivity. The geographic detector model can quantitatively analyze the weights of each driving factor and detect the extent of the interactions between multiple factors and the dependent variable [23]. In recent years, many scholars have employed the geographic detector method to conduct driving factor analyses of vegetation NPP in various regions [20,24,25].

Generally speaking, the current research on changes in vegetation NPP is mainly focused on the overall regional vegetation system or specific vegetation, and there is a lack of comparative research on multiple vegetation types in different ecosystem environments; the attribution analysis of changes in vegetation NPP is also primarily limited to the influence of climate factors (mainly precipitation and temperature) [26,27,28,29] and human activities (primarily land use) [30,31,32,33]. There are still certain research gaps on the effects of soil and elevation, population density, and gross domestic product (GDP) on vegetation NPP dynamics. Through comparative analysis of multiple vegetation types in different ecosystem environments, this paper can gain a deeper understanding of the different responses of vegetation ecosystems to driving factors of the natural environment and human activities. In addition, this article expands the driving factors of the study beyond common climate variables and human activity factors, taking into account various factors, including soil, elevation, population density, and GDP. The impact of driving factors on NPP dynamics was explored from the perspectives of spatial variation and temporal variation, and the five most important driving factors affecting the dynamic changes in the NPP of various vegetation to the north and south of the Hu Line were summarized and analyzed. This provides a more comprehensive perspective for analyzing driving factors of vegetation NPP changes. The main objectives of our study are as follows: (1) to determine the dynamic changes and driving factors of the NPP of nine vegetation types in the areas north and south of the Hu Huanyong Line in China from 2001 to 2022, (2) to deeply analyze the individual and combined driving effects of various factors on the dynamic changes in the NPP for each vegetation in the two regions and to determine the five main driving factors and the driving mechanism affecting the dynamics of the NPP for each vegetation, and (3) to provide insights into China’s ecological protection projects crossing the Hu Line based on the dynamic characteristics of the NPP of Chinese vegetation and by comparing the differences in the driving factors of NPP dynamic changes across different vegetation types in various regions. In general, the research in this article aims to deeply reveal the reasons for the formation of and change in different ecosystem patterns in China. It can provide a certain reference value for realizing regional ecological protection and sustainable development based on local conditions.

2. Materials and Methods

2.1. Overview of the Study Area

China experiences distinct seasons and has a complex and diverse climate. Generally, the southern regions have a warm and humid climate with an average annual temperature ranging from approximately 15 °C to 25 °C and an annual precipitation of around 800 mm to 2000 mm. In contrast, the northern areas are cold and dry, with an average annual temperature of between −5 °C to 10 °C and an annual precipitation of about 200 mm to 800 mm (Figure 1a,b). China’s topography and landforms are complex and diverse. Overall, the terrain is higher in the west and lower in the east, with mountains, plateaus, and hills accounting for approximately 67% of the total land area. Basins and plains comprise about 33% of the total land area (Figure 1c). The region has abundant rivers and lakes (Figure 1d). Influenced by the diversity of climate and topography, China exhibits various vegetation types, including coniferous forests, broad-leaved forests, grasslands, meadows, shrublands, cultivated vegetation, high mountain vegetation, wetlands, and deserts.

2.2. Data Sources

This paper collected data from 14 representative types under six major categories: ecology, climate, soil, socio-economics, vegetation, and topography. The specific data formats, sources, and preprocessing methods are presented in

Table 1. Data sources and processing methods.

Data Set		Time Resolution	Data Source
Type	Name
Ecology	NPP Data Set (2001–2022)	1 year	NASA’s MOD17A3HGF.061 product
Climate	Average Precipitation Data Set (2001–2021)	1 month	National Earth System Science Data Center (www.geodata.cn) (accessed on 20 October 2023)
	Average Temperature Data Set (2001–2021)
	Potential EvAPOtranspiration Data Set (2001–2021)
	Relative Humidity Data Set (2001–2020)
	Sunshine Duration Data Set (2001–2020)
Soil	Soil Type Data Set (1995)	1 year	Harmonized World Soil Database (HWSD)
	Soil pH Data Set (2010)
	Soil Moisture (30 cm) Data Set (2001–2020)	1 day	National Tibetan Plateau Scientific Data Center
Socio-economic	Land Use Data Sets (2000, 2005, 2010, 2015, 2020)	1 year
	GDP Raster Data Sets (2000, 2005, 2010, 2015, 2020)		Chen J. et al., 2022 [34]
	Population Density Data Set (2001–2021)		LandScan population dataset developed by the U.S. Department of Energy’s Oak Ridge National Laboratory
Vegetation type	Spatial Distribution Data Set of Vegetation Types in China (2000)	1 year	Resources and Environmental Science and Data Center, Chinese Academy of Sciences
Landform	China DEM Data Set (2000)	1 year

. For the convenience of analysis and research, the spatial resolution of all data is unified to 1 km × 1 km, and the projection coordinate system used is the China Lambert Conformal Conic projection coordinate system.

Vegetation-type source data were reclassified into nine categories: coniferous forest (CF, covering 9.02% of the total area), broadleaf forest (BF, covering 7.25% of the total area), scrub (covering 9.96% of the total area), tussock (covering 3.40% of the total area), grassland (covering 14.52% of the total area), meadow (covering 10.67% of the total area), alpine vegetation (AV, covering 3.44% of the total area), cultivated vegetation (CV, covering 22.31% of the total area), and marsh vegetation (covering 0.64% of the total area). Unclassified desert and water bodies were collectively categorized as the no-vegetation area (accounting for 18.80% of the total area; this article does not analyze the dynamic changes and driving factors of NPP in the no-vegetation area). Among these, 99.69% of the alpine vegetation was located north of the Hu Line, and 98.20% of the tussock vegetation was located south of the Hu Line. For the convenience of analysis, alpine vegetation is classified as vegetation in the northern area of the Hu Line, and tussock is classified as vegetation in the southern area (Figure 2a).

The paper’s most crucial original dataset is the multi-year NPP data for China, sourced from NASA’s MOD17A3HGF.061 product. The data are in TIFF format as raster data. After a series of preprocessing such as cropping the source data, multiplying by scaling factors, resampling, and averaging over many years, we can obtain China’s average NPP value from 2001 to 2022 in gC/(m²·a) (Figure 2b).

2.3. Research Methods

2.3.1. Linear Trend Analysis Method

The one-variable linear regression method based on the least squares principle is a commonly used approach for studying the spatiotemporal variation of vegetation ecological indicators [35,36]. This method is employed to conduct linear trend analysis on each pixel of the processed annual NPP raster dataset, with the calculation formula presented as follows in Equation (1):

$$S=\frac{n\times {\displaystyle \sum _{i=1}^{n}i\times V_i-{\displaystyle \sum _{i=1}^{n}i{\displaystyle \sum _{i=1}^n{V}_i}}}}{n\times {\displaystyle \sum _{i=1}^n{i}^2}-{({\displaystyle \sum _{i=1}^{n}i})}^2}$$

(1)

In the equation, n represents the number of years of data for calculation, which in this equation is 22; i represents the year number; V_i represents the NPP value for the i-th year; and S, denoted as Slope, represents the changing slope of the vegetation NPP trend line. When S > 0 for a specific pixel, it indicates an increasing trend in vegetation NPP on that pixel; conversely, when S < 0, it indicates a decreasing trend [35]. Pixels with S equal to 0 often correspond to areas without vegetation cover.

To better express the significance of linear trend line changes on this basis, this article uses the widely used F-test to test the significance of the trend line [37,38]. The formula is as follows in Equation (2):

$$\left\{\begin{array}{c}F=(n-2)\frac{U}{Q}\\ U={\displaystyle \sum _{i=1}^n{({\widehat{v}}_i-\overline{v})}^2}\\ Q={\displaystyle \sum _{i=1}^n{(v_i-{\widehat{v}}_i)}^2}\end{array}\right.$$

(2)

In the equation, the definitions of i and n are identical to those in Equation (1); F represents the F-test statistic; U stands for the sum of squares for regression with 1 degree of freedom; Q denotes the sum of squares for residuals with n − 2 degrees of freedom, which is 20 in this equation; v_i represents the value of variable v for the i-th year; ${\widehat{v}}_i$ represents the linear regression value of variable v for the i-th year; and $\overline{v}$ signifies the n-year average value of variable v.

Then, using the critical value relationship between the S- and F-values at various P-levels, the changes in the linear trend lines are categorized into seven types (

Table 2. Trend line change degree division.

S-Value	Significance Test Level p-Value	F-Value	Significance Change Type
$\in (0,+\infty )$	$\in (-\infty ,0.05)$	$\in (4.351,8.096)$	Significant increase (SI)
$\in (0,+\infty )$	$\in (-\infty ,0.01)$	$\in (8.096,14.819)$	More significant increase (MSI)
$\in (0,+\infty )$	$\in (-\infty ,0.001)$	$\in (14.819,+\infty )$	Extremely significant increase (ESI)
$\in (-\infty ,+\infty )$	$\in (0.05,+\infty )$	$\in \varnothing$$\mathrm{or}\in (-\infty ,4.351)$	No significant change
$\in (-\infty ,0)$	$\in (-\infty ,0.05)$	$\in (4.351,8.096)$	Significant decrease (SD)
$\in (-\infty ,0)$	$\in (-\infty ,0.01)$	$\in (8.096,14.819)$	More significant decrease (MSD)
$\in (-\infty ,0)$	$\in (-\infty ,0.001)$	$\in (14.819,+\infty )$	Extremely significant decrease (ESD)

).

2.3.2. Geodetector Model

The Geodetector is a set of statistical methods used to detect spatial heterogeneity and uncover the underlying driving forces [23]. The Geodetector comprises primarily four types of detector models, and this article primarily utilizes the following three detector models:

(1)

Factor detector

The factor detector mainly detects the explanatory power of a certain influence factor X to the spatial differentiation of the dependent variable Y. This explanatory power is measured by the q-value, which is expressed as follows in Equation (3):

$$q=1-\frac{{\displaystyle \sum _{i=1}^n{N}_i{\sigma }_i^2}}{N{\sigma }^2}=1-\frac{SSW}{SST}$$

(3)

In the equation, i = 1,2,3,…, n, represents the number of partitions or categories for variable Y or factor X; N_i and N respectively denote the number of units in zone i and the entire zone; and ${\sigma }_i^2$ and ${\sigma }^2$ represent the variance of Y-values in zone i and the whole zone, respectively. SSW represents the sum of within-zone variances, and SST represents the total variance of the entire zone. The range of q is [0, 1]. When q is 0, it means that factor X cannot explain the spatial heterogeneity of dependent variable Y at all, implying that factor X has no impact on Y; when q is 1, it indicates that factor X completely dominates the spatial distribution of dependent variable Y; and the higher the q value, the stronger the explanatory power of factor X for dependent variable Y.

(2)

Interaction detector

When exploring two or more influencing factors, the interaction detector can detect and calculate the degree of explanation (q-value) of the dependent variable Y when any two influencing factors X_j and X_k (j ≠ k) jointly affect it. By comparing the q-value after their interaction (q (X_j ∩ X_k)) with the q-values when they act individually (q (X_j), q (X_k)), the results of the interaction can be categorized into several types (

Table 3. Interaction type criteria.

q-Value Relation	Interaction Type
q (Xj∩Xk) < Min (q (Xj), q (Xk))	Nonlinearity attenuation
Min (q (Xj), q (Xk)) < q (Xj∩Xk) < Max (q (Xj), q (Xk))	The single-factor nonlinearity decreases
q (Xj∩Xk) > Max (q (Xj), q (Xk))	Two-factor enhancement
q (Xj∩Xk) = q (Xj) + q (Xk)	Independent
q (Xj∩Xk) > q (Xj) + q (Xk)	Nonlinear enhancement

):

In a system with multi-variable synergistic interactions, we can statistically calculate the mean value of q-value of the interaction results of a certain factor j and other factors $\overline{q_j}$ to represent the comprehensive dominance degree of a certain factor in the system, as shown in Equation (4).

$$\overline{q_j}=\frac{{\displaystyle \sum _{k=1}^{n}q(X_j\cap X_k)}}{n},j=1,2,\dots ,n$$

(4)

(3)

Risk detector

The outcomes of the risk detector can reveal the mean of the dependent variable on each partition of the respective variable X and determine whether there is a significant difference in the mean of the dependent variable between any two partitions of variable X.

2.3.3. Determination and Partition Methods of Driving Factors

As described earlier, when applying the Geodetector model for factor detection analysis, it is essential to complete three prerequisite steps: (1) determining the independent variable X and the dependent variable Y, (2) acquiring the values of variables X and Y, and (3) performing a rational partitioning of the independent variable factor X.

For (1), this study selected a total of 12 driving factors [39,40,41] from climate, soil, socio-economic, and topographic factors as independent variables X, including mean precipitation (mean P), mean temperature (mean T), relative humidity (RH), potential evapotranspiration (PET), sunshine duration (SD), soil type (ST), soil pH (SPH), soil moisture (SM), population density (PD), gross domestic product (GDP), land use type (LUT), land use/land cover change (LUCC), and elevation (DEM). NPP serves as the dependent variable Y. The meanings and units of each variable are provided in

Table 4. Factor description.

Variable Type	Variable Name	Variable Description	Variable Unit
Climate	Mean P	Average annual precipitation from 2001 to 2021	mm
	Mean T	Average annual temperature from 2001 to 2021	℃
	PET	Average annual potential evapotranspiration 2001–2021	mm
	RH	Average annual relative humidity from 2001 to 2020	%
	SD	Annual average sunshine duration 2001–2020	hr
Soil	SM	Annual average soil moisture from 2001 to 2020 (30 cm)	m3/m3
	SPH	Soil pH value in 2010	/
	ST	Soil types in 1995	/
Socio-economic	PD	Average annual population density 2001–2021	People/km2
	GDP	GDP in 2000, 2010, 2015, and 2020	trillion yuan
	LUT	Land use types in 2000, 2010, 2015, and 2020	/
	LUCC	Land use/land cover-type changes in 2000, 2010, 2015, and 2020	/
Landform	DEM	Digital elevation model	m

.

For (2), due to the large study area, obtaining values for each pixel is not practical. Therefore, when conducting a Geodetector analysis, the values of variables X and Y are extracted by creating many random sampling points. In this paper, the annual mean of each factor is extracted as the spatial variation. The multi-year linear change trend S-value is used as temporal variation to analyze the driving effect of each factor’s spatial and temporal change on the NPP dynamic change of vegetation.

For (3), the partitioning of the independent variable X for each factor is shown in

Table 5. Partition method for argument X.

Factor Type	Factor Name	X
		Mean		Slope Value
		Partition Method	Number of Partitions	Partition Method	Number of Partitions
Climate	Mean precipitation	Jenks natural breakpoint method	6	Jenks natural breakpoint method	8
	Mean temperature		6		8
	Relative humidity		8		8
	Potential evapotranspiration		6		8
	Sunshine duration		8		8
Soil	Soil type	Partition by definition	13	\	\
	Soil pH	Jenks natural breakpoint method	6	\	\
	Soil moisture		8	Jenks natural breakpoint method	8
Socio-economic	Population density	Quantile classification	10	Quantile classification	10
	GDP		10		10
	Land use	Partition by definition	8	Partition by definition	43 ①
Landform	DEM	Jenks natural breakpoint method	8	\	\

Note: “\” indicates factors that do not possess or are not considered for multi-year change trends. ①: The conversion of land use types for 2000 and 2020 is subjected to visual transformation analysis. This analysis is conducted based on the first-level categorization into 7 classes, resulting in 6 × 7 = 42 distinct combinations of changes. Additionally, the transformation within the same category is classified as “no change”, resulting in 43 categories.

. Among them, soil type is digitized based on the “1:1,000,000 Soil Map of the People’s Republic of China” (1995) and reclassified into 13 categories based on data (Figure 3a), corresponding to 13 partitions; land use types are classified into 7 categories according to the first-level land classes (Figure 3b).

3. Results

3.1. Dynamic Characteristics of NPP in Vegetation on Both Sides of the Hu Line

Through spatial analysis, we got the change slope of the overall NPP of vegetation (Figure 4a), the significance of change (Figure 4b), and the annual changes in the NPP of various vegetation in China (Figure 5). It can be observed that (1) regions in China exhibiting an increasing trend in multi-year vegetation NPP are primarily located in the Loess Plateau regions of Gansu, Shanxi, and Shaanxi provinces, as well as the Sichuan Basin and the Northeast Plain. These areas are concentrated around the Hu line. Areas with decreasing NPP trends are mainly found in the southeastern parts of China, including provinces such as Jiangsu, Fujian, Jiangxi, Guangdong, Hainan, and Taiwan, as well as the southern regions of Yunnan Province and the border areas of southern Tibet. (2) In the two areas north and south of the Hu Line, the predominant trend in vegetation NPP changes for different vegetation types is either an increase or a slow increase. Among these, north of the Hu Line, the swamp vegetation exhibits the highest NPP increase trend (4.048 gC/(m²·a²)), followed by cultivated vegetation (3.490 gC/(m²·a²)), and alpine vegetation has the most miniature increase trend (0.360 gC/(m²·a²)). South of the Hu Line, grassland has the most significant NPP change trend (5.320 gC/(m²·a²)), followed by meadow (4.087 gC/(m²·a²)), and swamp vegetation shows the smallest change trend (1.787 gC/(m²·a²)).

According to the distribution of significant areas of NPP changes for the nine vegetation types (Figure 6), it is evident that on the north side of the Hu Line, the area in which the NPP of grassland shows an increasing trend (including SI, MSI, and ESI) is the largest and contributes the most to the increase in NPP, covering 52.58 × 10⁴ km². On the south side of the Hu Line, the area where the NPP of cultivated vegetation shows an increasing trend is the largest and contributes the most to the increase in NPP, covering 103.50 × 10⁴ km².

There is a total of 20.95 × 10⁴ km² of vegetated areas where the NPP exhibits a decreasing trend (including SD, MSD, and ESD). Among these, cultivated vegetation, coniferous forest, broadleaf forest, and scrubland account for 29.12%, 21.00%, 14.97%, and 15.29%, respectively, making them the most significant vegetation types with decreasing NPP trends.

3.2. Identification of Driving Factors of NPP Dynamic Change in Vegetation

3.2.1. The Impact of Spatial Variation of Factors on Dynamic Changes in NPP

Utilizing the Geodetector model, we can obtain the factor detection results (q-values) for the influence of each factor’s spatial variation on the dynamic change trend of the NPP in vegetation types on both sides of the Hu Line (Figure 7), as well as the interaction detection results (Figure 8). Based on the interaction detection results, the comprehensive explanatory power of each factor is calculated. Combining this with the risk detection results, the driving effects of the spatial variations of five main driving factors on the dynamic changes in NPP in vegetation on both sides of the Hu Line are identified (Figure 9). The specific results of the driving effect of the spatial variations of these five main driving factors on the dynamic changes in the NPP in vegetation north and south of the Hu Line are detailed in Figure 10.

Since the article involves many vegetation types and driving factors and divides the research area into two parts for research, to discuss the results as concisely and comprehensively as possible, several abbreviations are listed here: (1) “Dynamic changes in NPP” is abbreviated as “NPP_d”; (2) “spatial variation in driving factors” is abbreviated as “X_s”, where X is the abbreviation of the name of any driving factor, such as the spatial variation of DEM, abbreviated as DEM_s; and (3) “temporal variation of driving factor” is abbreviated as “X_t”, where X is the abbreviation of the name of any driving factor, such as the temporal variation of mean P, abbreviated as “mean P_t”.

In general, the single-factor explanatory power of the spatial variation of each factor on the NPP_d of vegetation in the northern area of Hu Line is greater than that in the southern area (q (N) = 0~0.496; q (S) = 0~0.207). The spatial variation of natural environmental factors (climate, soil, and elevation) dominates the NPP_d of most vegetation, while the explanatory power of human activity factors (PD, GDP, and LUT) is generally not significant. The common explanatory power of every two factors enhances the explanatory power of every single factor, and the common driving effect of climate factors and elevation is the most obvious.

Specifically, (1) in the area north of the Hu Line, the NPP_d of CF is most driven by DEM_s (q = 0.307), and the joint effect of DEM_s and the spatial variation of each climate factor has a significant impact on the NPP_d of CF, among which DEM_s∩RH_s has the most significant explanatory power (q = 0.391). The NPP_d of BF is largely driven by the spatial variation of climate and soil factors (q > 0.3). The explanatory power of GDP_s on the NPP_d of BF reaches 0.404, and BF is also the only vegetation type whose NPP_d is significantly affected by GDP_s. The NPP_d of grassland is most driven by ST_s, SPH_s, and DEM_s (q = 0.342, 0.333, and 0.391, respectively). The joint effect of DEM_s and other factors significantly impacts the NPP_d of grassland, among which DEM_s∩Mean P_s has the largest explanatory power, with q reaching 0.602. In addition, one point that cannot be ignored is that PD_s has the greatest explanatory power for the NPP_d of grassland among all vegetation (q = 0.209), and the explanatory power q of the two-factor combination of PD_s and DEM_s on the NPP_d of grassland reaches 0.535. The NPP_d of CV is mainly driven by the mean P_s and ST_s (q = 0.378, 0.329). The NPP_d of marsh is primarily driven by the mean T_s, SD_s, and DEM_s (q = 0.248, 0.235, 0.253). The single-factor driving effects of the spatial variation of each factor on scrub, meadow, and AV are all small (q < 0.2), but the joint effect of DEM_s and the spatial variation of each climate factor has greater explanatory power on the NPP_d of meadow (q = 0.260~0.336). In the area south of the Hu Line, the single-factor explanatory power of the spatial variation of each driving factor on the NPP_d of each vegetation type is small, and only mean T_s and SD_s have driving explanatory power q > 0.2 for the NPP_d of grassland, among which DEM_s∩SD_s has the largest explanatory power (q = 0.352). In addition, the joint effect of some double factors has greater explanatory power on the NPP_d of meadow: DEM∩RH (q = 0.336), DEM∩SD (q = 0.309), and RH∩SD (q = 0.301).

According to the five most crucial comprehensive driving factors of each vegetation type, it can be found that there are differences in the types of driving factors and the magnitude of the driving forces for the same vegetation type in different natural geographical environments (the area south of the Hu Line and the area north of the Hu Line). In addition, it is worth emphasizing that although the overall explanatory power is not large, only the five main driver species of CV on the south side of the Hu Line have human activity category factors (LUT).

3.2.2. The Impact of Temporal Variation of Factors on Dynamic Changes in NPP

After analysis, the factor detection results for the influence of each factor’s temporal variation on the dynamic change in NPP in various vegetation types on both sides of the north and south are obtained (Figure 11), as are the interaction detection results (Figure 12). Based on the factor and interaction detection results, the influence of the temporal changes in five main driving factors on the dynamic changes in vegetation NPP on both sides of the Hu Line is analyzed (Figure 13). Further analysis based on these results yields the driving relationship results of the temporal variation of these five main driving factors on the dynamic changes in vegetation NPP on both the north and south sides of the Hu Line (Figure 14).

In general, similar to the impact of the spatial variation of factors on the NPP_d of vegetation, the temporal variation of each factor has a greater single-factor driving force for each factor in the northern region of the Hu Line than in the southern region (q (N) = 0~ 0.499; q (S) = 0~0.230). However, the explanatory power of each factor has undergone some changes. The temporal variation of natural environmental factors is still the dominant reason driving most NPP_d of vegetation and the driving force of human activity factors is still generally not significant. The common explanatory power of every two factors enhances the explanatory power of every single factor, and the common driving effect of RH and other climate factors is the most obvious.

Specifically, (1) in the area north of the Hu Line, PET_t has the greatest driving effect on the NPP_d of CF (q = 0.305), and the joint driving effect of PET_t and the temporal variation of other climate factors has a significant impact on the NPP_d of CF. The NPP_d of BF is mainly driven by the single factors of mean P_t, RH_t, PET_t, and SD_t (q = 0.499, 0.371, 0.451, and 0.440, respectively). The NPP_d of grassland is mainly driven by mean P_t and RH_t (q = 0.368, 0.212), and the explanatory power q of the joint effect of RH_t∩Mean P_t reaches 0.503. In addition, for grassland, although the explanatory power of PD_t and LUCC for its NPP_d is not significant (q = 0.101, 0.137), this value is much greater than the explanatory power of the NPP_d of PD_t and LUCC on other vegetation types. In addition, the explanatory power q of the two-factor combination of PD_t and mean P_t and the two-factor combination of LUCC and mean P_t reach 0.403 and 0.458, respectively, for the NPP_d of grassland. The NPP_d of CV is mainly driven by mean P_t, RH_t, PET_t, and SD_t (q = 0.338, 0.290, 0.287, and 0.283, respectively), among which RH_t∩Mean P_t has the greatest explanatory power (q = 0.446). The single-factor driving effects of the temporal variation of each factor on scrub, meadow, AV, and marsh are all small (q < 0.2), but among them, the combined effect of the temporal variation of mean T_t and other climate factors has a greater driving force on the NPP_d of meadow (0.2 < q < 0.3), the joint effect of the temporal variation of every two climate factors has a large driving force on the NPP_d of marsh (0.2 < q < 0.3), and the joint effect of LUCC and the temporal variation of each climate factor also has certain explanatory power on its NPP_d. (2) In the area south of the Hu Line, the temporal variation of each driving factor has a small single-factor driving effect on the NPP_d of each vegetation type. Only the driving explanatory power of RH_t on grassland is q > 0.2, among which the explanatory power q of RH_t∩Mean P_t reaches 0.436. In addition, the joint driving effects of some dual factors have greater explanatory power for the NPP_d of marsh: SD_t∩LUCC (q = 0.298), SD_t∩RH_t (q = 0.293), and SD_t∩PET_t (q = 0.291).

4. Discussion, Implications, and Limitations

4.1. Discussion

From 2001 to 2022, 38.22% of the vegetation coverage areas in China showed an increasing trend for NPP. Among them, the regions showing a relatively significant increase (MSI) and an extremely significant increase (ESI) are mainly located in the Northeast Plain, the Greater Khingan Range, the Inner Mongolia Plateau, the Loess Plateau, the Hexi Corridor, the central Qinghai Plateau, and the Sichuan Basin. Grassland and cultivated vegetation, which are widely distributed in these regions, are the two main types of vegetation showing an increasing trend in NPP. They contribute the most to the increase in NPP to the north and south of the Hu Line, respectively. This can be attributed to positive changes in natural factors such as climate in the region in recent years, as well as a series of ecological restoration and vegetation greening policies implemented by China in the area, such as “grain for green” (returning farmland to forests/grasses) [42,43,44], the “Three-North Shelterbelt Program” [45], and “natural forest resource protection” [46]. By quantifying the benefits of these policies in the form of NPP, we can intuitively see China’s “Green Breakthrough Hu Line” efforts in recent years. The regions showing a decreasing trend in NPP mainly include parts of the southeastern areas, including Jiangsu Province, Fujian Province, Jiangxi Province, Guangdong Province, Hainan Province, and Taiwan Province, as well as the border regions in the southern part of Yunnan Province and the southern part of Tibet, primarily due to the intensification of industrialization and urbanization processes in southwestern and coastal regions in recent years [47].

For the entire study area, the spatiotemporal changes in various factors have a generally greater driving effect on the dynamic changes in vegetation NPP in the areas north of the Hu Line than in the areas south. This shows that the dynamics of vegetation NPP in the area north of the Hu Line are generally more sensitive to the driving response of various factors. In contrast, the dynamics of vegetation NPP in the area south of the Hu Line are more stable in response to the driving factors of various factors. This is because the climate pattern in the area north of the Hu Line in China is generally relatively singular, the ecosystem is very fragile, and the vegetation structure and function will respond more obviously to changes in factors [22].

According to the detection results of driving factors, it can be found that climate factors, especially spatial changes in precipitation or temperature, play an important role in the dynamic changes of NPP of most vegetation (CF (N) BF, grassland, CV (N), and marsh), which is consistent with the conclusions of Li et al., Chen et al., and Hou et al. [48,49,50]. In addition, elevation and soil factors also greatly influence the dynamic changes in NPP of some vegetation. For example, the spatial variation of DEM has the greatest driving effect on CF (N), grassland (N), and marsh (N) (q = 0.307, 0.391, and 0.253, respectively). The influence of ST on BF (N), grassland (N), and CV (N) is significant (q = 0.328, 0.342, and 0.329, respectively). Moreover, the joint effect of DEM and the spatial variation of various climate factors has an obvious driving effect on the dynamic changes in NPP of various vegetation. This is mainly because elevation changes directly affect the light and heat environment, moisture, and soil physical and chemical properties for plant growth, and then affect its structure and function [51]. This shows that some scholars only consider climate factors to be natural driving factors of NPP changes, which lacks a certain degree of comprehensiveness [27,49].

From the perspective of human activity factors, the dynamic change in the broad-leaved forest on the north side of the Hu Line significantly responds to the spatial variation of GDP (q = 0.404). Combined with the distribution region of broad-leaved forest, it can be seen that the broad-leaved forest on the north side of the Hu Line is concentrated in southern Tibet, China. Southern Tibet is densely populated, and in recent years, the economy based on tourism has developed rapidly, which greatly impacts the ecological environment [52]. The dynamic changes in NPP in grassland on the north side of the Hu Line are greatly affected by human activity factors. In particular, the joint effect of the spatial variation of PD and DEM, the joint effect of the temporal variation of mean T, and LUCC significantly drive the dynamic changes in NPP (q = 0.535 and 0.458, respectively). This is because the grassland vegetation on the northwest side of the Hu Line is mainly desert grassland and pasture grassland. The desert grassland habitat is fragile and susceptible to external factors; the pasture grassland is mainly affected by human intervention. The dynamic changes in NPP of marsh vegetation on the north and south sides of the Hu Line are significantly affected by SD and LUCC. This is because marshes, as fragile ecosystems, are greatly affected by light and heat conditions and are more susceptible to the impact of human activities [53].

The explanatory power of the interaction between human activity and natural activity factors is greater than the explanatory power of the two alone, indicating that natural factors and human factors promote each other and cooperatively control the dynamic changes in NPP of each vegetation type.

4.2. Implications and Suggestions

This paper conducted an in-depth analysis of the NPP dynamics and driving factors of various vegetation types on both sides of the Hu Huanyong Line and found large differences in the size, changing trend, main driving factors, and driving effects of each vegetation NPP in different areas. Therefore, regional differences should be fully considered in formulating relevant policies such as ecological restoration and protection, and a more directional coordinated regional development of the country’s ecological environment should be achieved. Specifically, grassland vegetation contributes greatly to the increase in NPP for the ecologically fragile areas north of the Hu Line. Still, it is greatly affected by climate, elevation, and human activity factors. Therefore, ecological protection projects for grasslands should be implemented in a planned manner. As for broad-leaved forests and marsh vegetation, attention should also be paid to their protection in economic and social development. In addition, this paper found that on the north side of Hu Line, in addition to meadows and alpine vegetation mainly distributed in plateau areas, which are less affected by multiple driving factors, scrub vegetation also has great stability in the influence of various factors. Therefore, it is recommended that when carrying out preliminary greening work in the northwest, shrubbery vegetation should be planted first and then other vegetation types should be planted after the ecological environment improves.

Affected by the natural endowment of altitude, the “ecological barrier” of the Hu Huanyong Line cannot be eliminated, but this does not mean that it cannot be crossed from certain areas. Combined with the spatial and temporal changes in China’s NPP, it can be found that the northwest region is not all deserts that are difficult to develop. There are also some green areas with a good ecological environment in the form of points or lines, such as the Lanzhou–Xining area, the Ningxia Yellow River area, the Gansu Hexi Corridor area, the Urumqi area on the northern slope of the Tianshan Mountains, and the southeastern Tibet area. These areas are relatively rich in natural resources and have a good ecological environment. Priority can be given to expanding the ecological greening of this area, thereby promoting economic development and making it grow into an urban agglomeration, improving population agglomeration capacity, connecting points to lines, forming lines into planes, and ultimately achieving further crossing of the Hu Huanyong Line.

4.3. Limitations and Prospects

This paper uses the Geodetector model to analyze the dynamic driving factors of vegetation NPP. This model does not require any preset model form, can intuitively measure each factor’s degree and mechanism of influence on the research object, and can determine every two joint driving effects of factors. However, the accuracy of the results has extensive requirements for the quantity and quality of data and is greatly affected by variable selection and variable partitioning methods [54]. In subsequent research, we can consider combining geographical detectors with other driving factor identification methods such as decision trees [55,56], random forests [57,58], and geographically weighted regression (GWR) [59,60] for mutual verification to ensure the reliability of the results further. In addition, compared with the three human activity factors selected in this article (GDP, population density, and land use), to further highlight the impact of human activity factors on vegetation NPP dynamics, it may be considered to add some factors that better reflect the intensity of human activities for analysis, such as human activity footprint, resource consumption intensity, and environmental pollution level [61,62].

5. Conclusions

With the change in global climate and the intensification of human activities, the structure and function of terrestrial ecosystems have undergone tremendous changes in recent years. Most studies have shown that China’s land cover has become significantly greener in recent years, but this greening trend is not effective when crossing the Hu Huanyong Line. Therefore, this study conducted a comprehensive and targeted analysis of the dynamic changes and driving factors of the net primary productivity of nine vegetation types in two areas separated by the Hu Line. The research conclusions are as follows:

(1)

Over the past 20 years, 38.22% of the regional vegetation NPP in China increased, mainly in the Loess Plateau, Sichuan Basin, and Northeast Plains, while 2.39% decreased, primarily in southeastern China and southern Tibet. The NPP for all vegetation types increased. Grasslands contributed the most to NPP growth north of the Hu Line (39.71%), while cultivated vegetation was the primary contributor south of the Hu Line (50.58%).

(2)

The spatiotemporal changes in various factors have a generally more significant driving effect on the dynamic changes in vegetation NPP in the areas north of the Hu Line than in the regions south. The vegetation ecosystem on the north side of the Hu Line is more fragile than that on the south side.

(3)

The grassland and broad-leaved forest on the north side of the Hu Line, as well as the marshes on the north and south sides, are greatly affected by human activities. The dynamic changes in NPP of other vegetation types are mainly affected by natural activities, with the most significant effect being the combined effect of elevation and climate factors. Shrubs, alpine vegetation, and meadows show minimal response to the driving impact of individual factors (q < 0.2).