Spatiotemporal Evolution and Driving Factors of NPP in the LanXi Urban Agglomeration from 2000 to 2023

Abstract

This study quantitatively evaluates the effects of human activities (HAs) and climate change (CC) on the terrestrial ecosystem carbon cycle, providing a scientific basis for ecosystem management and the formulation of sustainable development policies in urban agglomerations located in arid and ecotone regions. Using the LanXi urban agglomeration in China as a case study, we simulated the spatiotemporal variation of vegetation net primary productivity (NPP) from 2000 to 2023 based on MODIS remote sensing data and the CASA model. Trend analysis and the Hurst index were employed to identify the dynamic trends and persistence of NPP. Furthermore, the Geographical Detector model with optimized parameters, along with nonlinear residual analysis, was employed to investigate the driving mechanisms and relative contributions of HAs and CC to NPP variation. The results indicate that NPP in the LanXi urban agglomeration exhibited a fluctuating upward trend, with an average annual increase of 4.26 gC/m² per year. Spatially, this trend followed a pattern of “higher in the center, lower in the east and west,” with more than 95% of the region showing an increase in NPP. Precipitation, mean annual temperature, evapotranspiration, and land use types were identified as the primary driving factors of NPP change. The interaction among these factors demonstrated a stronger explanatory power through factor coupling. Compared with linear residual analysis, the nonlinear model showed clear advantages, indicating that vegetation NPP in the LanXi urban agglomeration was jointly influenced by HAs and CC. These findings can further act as a basis for resource and environmental research in similar ecotone regions globally, such as Central Asia, the Mediterranean Basin, the southwestern United States, and North Africa.

1. Introduction

Vegetation represents a crucial part of terrestrial ecosystems, playing a crucial role in mitigating extreme climate events, maintaining carbon cycles, and ensuring ecosystem stability [1]. With the progression of global informatization and industrialization, the impacts of human activities (HAs) and climate change (CC) on vegetation distribution and biomass have become increasingly profound [2]. Vegetation net primary productivity (NPP) has been widely utilized in research related to vegetation growth [3]. It denotes the amount of organic matter amassed per unit area and unit time by green plants through photosynthesis by converting light energy into chemical energy. Specifically, NPP signifies the net amount of carbon fixed by plants from atmospheric CO₂ through photosynthesis [4,5]. As a critical indicator of material and energy cycling in ecosystems, vegetation NPP reflects both ecosystem sustainability and the joint effects of HAs on the terrestrial biosphere [6]. It provides a direct representation of how HAs and CC influence vegetation. Therefore, studying the spatiotemporal evolution and driving mechanisms of vegetation NPP is essential for monitoring ecological conditions and assessing the sustainability of ecosystems. In quantitative NPP research, several estimation models have been proposed, including climate productivity models, physiological ecology models, and light-use efficiency models [7,8]. Among them, the Carnegie–Ames–Stanford Approach (CASA), based on light-use efficiency principles, stands out for its simplicity, lack of fieldwork requirements, and compatibility with remote sensing data. The use of remote sensing to estimate vegetation NPP has been continuously refined and widely adopted [9], particularly for analyzing the spatiotemporal dynamics of NPP.

For exploring spatiotemporal trends and evolutionary patterns, trend and stability analyses based on long-time-series data are highly effective in uncovering the internal mechanisms of vegetation NPP. These methods are both stable and reliable and have been widely applied in international [10] and domestic [11] research. At the spatial-scale level, national- [12], provincial- [13], and watershed-scale [14] studies are relatively comprehensive; however, research focused on vegetation NPP in urban agglomerations remains limited. Urban agglomerations serve dual roles in urban development and ecological protection, making them critical areas for vegetation NPP studies. Vegetation NPP is typically influenced by multiple factors, primarily HAs and CC [15]. Climate variables are considered the dominant drivers of vegetation dynamics, while anthropogenic activities also play a significant role. For instance, ecological restoration policies, such as returning farmland to forest, have effectively enhanced vegetation NPP [16]. Conversely, land use type (LU) changes and large-scale urban expansion driven by human activity can reduce NPP. Hence, future ecological decision making and policy development must consider the dual effects of HAs and CC, allowing for more targeted restoration and protection strategies that promote regional ecosystem sustainability.

Previous studies have commonly used linear regression [17], partial correlation analysis [18], and similar methods to identify factors affecting NPP distribution. However, these approaches often overlook the spatial heterogeneity of NPP, which arises from the interaction of multiple drivers. Traditional linear models are limited in their ability to capture such complex relationships. To address this issue, Wang et al. [19] introduced the Geographical Detector method, which quantifies the influence of multiple factors on vegetation NPP and identifies interactions between them. The optimized parameter version of the geographical detector has since gained wide acceptance [20], and it is now an important tool for identifying key explanatory variables from a range of potential drivers. Another key challenge is to quantify the relative contributions of natural and anthropogenic factors to changes in NPP. Some researchers have used linear residual models to assess these contributions by establishing statistical links between NPP and its drivers [21]. However, due to the complexity of interactions between natural variability and human activity, linear methods alone often fall short. Therefore, this study compares both linear and nonlinear residual trend analysis approaches to more accurately evaluate the relative impacts of HAs and CC on vegetation NPP.

Du [22] pointed out that urban agglomerations are regions where HAs are particularly frequent and intense, while Li [23] emphasized that the role of urbanization as a driving factor of NPP remains an unresolved issue. In ecologically fragile regions, urban agglomerations exhibit especially complex human–environment interactions, which have a profound impact on urban sustainability. The NPP of urban agglomerations is shaped by the coupled effects of ecological–climatic conditions and HAs. As a result, this topic has drawn increasing global scholarly attention, underscoring the significance of conducting research in this area.

The Lanzhou–Xining urban agglomeration (hereinafter referred to as the LanXi urban agglomeration), located at the transitional interface between the Qinghai–Tibet Plateau and the Loess Plateau, serves as both a vital component of China’s northwestern ecological barrier and a core area within the country’s “Belt and Road” strategic initiative. The region is marked by a dual identity of ecological fragility and strategic economic importance. Its distinctive location, high ecological sensitivity, and intense human activity make it exceptionally representative and practically significant for studying the spatiotemporal evolution of vegetation NPP. Accordingly, the LanXi urban agglomeration provides an ideal setting for examining the coupled effects of natural environmental changes and anthropogenic influences. This study uses the LanXi urban agglomeration as a typical case and investigates the spatiotemporal dynamics of NPP from 2000 to 2023. It employs an integrated methodological framework, including the CASA model, GIS-based spatial analysis, Sen’s slope estimator, the Hurst index, the optimal parameter Geographical detector (OPGD), and grid-based residual analysis. These tools are applied to systematically explore the driving mechanisms of HAs and CC on regional NPP, to reveal the spatial patterns and persistence of NPP changes, and to deeply analyze the interactive relationships and relative contributions of natural and anthropogenic factors.

The findings of this research not only uncover the ecosystem’s response patterns to CC in this region but also quantify the influence of urbanization and economic development on NPP. These insights provide a scientific foundation for constructing regional ecological security frameworks, optimizing land resource allocation, and formulating sustainable development policies. Focusing on the ecological transition zone represented by the LanXi urban agglomeration, this study aims to analyze the spatiotemporal evolution of NPP from 2000 to 2023, assess trends and stability in vegetation NPP, evaluate the influences of HAs, CC, and their interactions on the spatial distribution of NPP, and quantitatively determine the relative contributions of HAs and CC to NPP changes. This research ensures the timeliness and relevance of its findings through a multi-source, data-driven framework. By distinguishing the relative influence of climatic and anthropogenic drivers, it contributes a valuable scientific basis for developing effective strategies for ecosystem protection and restoration. Furthermore, the study offers meaningful references for similar NPP research in ecological transition zones worldwide.

2. Materials and Methods

2.1. Study Area

The region exhibits significant topographic variation. As shown in Figure 1a, it lies at the junction of diverse landforms, extending from the Loess Hills in the east to plateau basins in the west. Figure 1b illustrates the region’s broad elevation range and complex terrain, which supports a mosaic of ecosystems, including forests, grasslands, croplands, and water bodies (Figure 1c). Additionally, this area is situated within a transitional zone between a temperate monsoon climate and a plateau climate, and it is characterized by uneven spatiotemporal precipitation distribution. The ecosystem is highly sensitive to CC, with vegetation growth conditions and NPP displaying marked spatial heterogeneity and temporal variability. Figure 1d highlights that the LanXi urban agglomeration is a key hub of urbanization and industrialization in Northwest China. With rising population density (POP) and accelerating urban expansion, LU in the region has undergone dramatic transformations: urban construction land has rapidly increased, while cropland and grassland have been progressively encroached upon, leading to evident local ecological degradation. Consequently, intensified human activity has had an increasingly significant impact on NPP. This region serves as a typical example of the tension and potential synergy between ecological protection and economic development. Important nature reserves, such as Qinghai Lake, play a vital role in regional ecological barrier construction, while Lanzhou and Xining, as major urban centers, bear critical responsibilities for economic development. The LanXi urban agglomeration is thus a representative ecological transition zone, where strong coupling between natural environmental change and HAs underscores its strategic importance for research on sustainable vegetation NPP development.

2.2. Data

The dataset used in this study includes vegetation NPP, LU, meteorological, elevation, socioeconomic, and related auxiliary data covering the period from 2000 to 2023. To ensure spatial consistency, all spatial datasets were uniformly projected using the Albers equal-area coordinate system. Detailed information on data sources and specifications is provided in

Table 1. Data sources and descriptions.

Data Type	Data Source	Data Description
LU type	Chinese Academy of Sciences Resources Science Data Center (https://www.resdc.cn/, accessed on 5 April 2025)	2000–2023
Soil data	World Soil Database (HWSD)	2020
Dem data	Geospatial Data Cloud (http://www.gscloud.cn/, accessed on 5 April 2025)	2020
Meteorological data	Tibetan Plateau Data Science Center (https://data.tpdc.ac.cn/, accessed on 5 April 2025)	2000–2023
Socio-economic density	Chinese Academy of Sciences Resources Science Data Center (https://www.resdc.cn/, accessed on 5 April 2025)	2000–2023
Basic geographic data	National Geomatics Center of China (https://www.ngcc.cn, accessed on 5 April 2025)	2020
Vegetation NPP	MOD17A3HGF (https://lpdaac.usgs.gov, accessed on 5 April 2025)	2000–2023

. Additionally, all raster datasets were resampled to a uniform spatial resolution of 500 × 500 m.

2.3. Methods

This study investigates the spatiotemporal variations and driving factors of vegetation NPP in the LanXi urban agglomeration from 2000 to 2023, following four sequential methodological steps, as illustrated in the research framework in Figure 2. First, relevant data were collected and processed. Second, a comprehensive spatiotemporal analysis of vegetation NPP was performed. Third, future trends and the stability of NPP were assessed. Finally, the impacts of HAs and CC on NPP were quantified using the OPGD model and nonlinear residual analysis.

2.3.1. CASA Model

Vegetation NPP reflects the carbon sequestration service function [3], reflecting the regulatory effect on climate. The CASA model [3] is adopted for the NPP calculation, reflecting the regulatory effect on climate:

NPP(x, t) = APAR(x, t) × ε(x, t)

(1)

where NPP(x, t) represents the carbon sequestration amount of the pixel (g/m²), APAR(x, t) represents the photosynthetically active radiation of pixel x in the t-th month (MJ/m²), and ε(x, t) represents the actual light energy utilization rate of pixel x in the t-th month (gC/MJ).

2.3.2. Sen Trend Analysis and Mann–Kendall Test

Since the temporal variation of vegetation NPP is a long-term, continuous, and complex coupled dynamic process, it is essential to analyze its trends and stability using time-series data. The Sen trend analysis, based on a univariate linear regression method, is employed to identify the changing characteristics of vegetation NPP [24]. In this method, the slope represents the rate of change in vegetation NPP, where xᵢ denotes the value of NPP in the i-th year, and i represents the year. A slope greater than zero indicates an increasing trend in vegetation NPP in the LanXi urban agglomeration, while a slope less than zero suggests a declining trend. Additionally, the slope results are categorized based on statistical significance: significant changes (p < 0.05) and non-significant changes (p > 0.05).

$$\mathrm{slope}={\displaystyle \frac{\mathrm{n}*{\sum }_{\mathrm{i}=1}^{\mathrm{n}}\left({\mathrm{i}*\mathrm{x}}_{\mathrm{i}}\right)-\left({\sum }_{\mathrm{i}=1}^{\mathrm{n}}\mathrm{i}\right)*({\sum }_{\mathrm{i}=1}^{\mathrm{n}}{\mathrm{x}}_{\mathrm{i}})}{\mathrm{n}*{\sum }_{\mathrm{i}=1}^{\mathrm{n}}{\mathrm{i}}^2-({\sum }_{\mathrm{i}=1}^{\mathrm{n}}{\mathrm{i})}^2}}$$

(2)

The Mann–Kendall test is a non-parametric statistical method commonly used to assess the significance of trends in time-series data. The test statistic S is calculated using Equation (3). Based on the corresponding significance level (p-value) of the Mann–Kendall statistic, the test determines whether the observed trend in the time series is statistically significant.

$$\mathrm{S}={\sum }_{\mathrm{i}=1}^{\mathrm{n}-1}{\sum }_{\mathrm{j}=\mathrm{i}+1}^{\mathrm{n}}\mathrm{sgn}\left({\mathrm{X}}_{\mathrm{j}}-{\mathrm{X}}_{\mathrm{i}}\right)$$

(3)

2.3.3. Hurst Index

The Hurst index has been widely used by researchers to analyze future change trends in long-term vegetation time series [24]. Consider the time series $\left\{{\mathrm{NPP}}_{\mathrm{H}\left(\mathrm{t}\right)},\mathrm{t}=1,2,\dots ,\mathrm{n}\right\}$. For any positive integer $\tau \ge 1$, we define the following: the mean sequence (Equation (4)), cumulative deviation (Equation (5)), range (Equation (6)), and standard deviation (Equation (7)).

$${\mathrm{NPP}}_{\mathrm{H}\left(\tau \right)}={\displaystyle \frac{1}{\tau }}{\sum }_{\mathrm{t}=1}^{\tau }{\mathrm{NPP}}_{\mathrm{H}\left(\mathrm{t}\right)},\tau =1,2,\dots ,\mathrm{n}\left(23\right)$$

(4)

$$\mathrm{X}\left(\mathrm{t},\tau \right)={\sum }_{\tau }^{\mathrm{t}}\left({\mathrm{NPP}}_{\mathrm{H}\left(\mathrm{u}\right)}-{\mathrm{NPP}}_{\mathrm{H}\left(\tau \right)}\right),1\le \mathrm{t}\le \tau$$

(5)

$$\mathrm{R}\left(\tau \right)=\underset{1\le \mathrm{t}\le \tau }{\max }\mathrm{X}\left(\mathrm{t},\tau \right)-\underset{1\le \mathrm{t}\le \tau }{\min }\mathrm{X}\left(\mathrm{t},\tau \right)$$

(6)

$$\mathrm{S}\left(\tau \right)={\left[{\displaystyle \frac{1}{\tau }}{\sum }_{\mathrm{u}=1}^{\mathrm{t}}{\left({\mathrm{NPP}}_{\mathrm{H}\left(\mathrm{t}\right)}-{\mathrm{NPP}}_{\mathrm{H}\left(\tau \right)}\right)}^2\right]}^{{\displaystyle \frac{1}{2}}}$$

(7)

The Hurst index ranges between 0 and 1. When H < 0.5H, it indicates that the future trend of the NPP time series is likely to reverse compared to the past, demonstrating anti-persistence. When H = 0.5H, the changes in vegetation NPP are random, showing no correlation with past trends. When H > 0.5H, the future trend tends to follow the past, indicating persistence.

2.3.4. OPGD Model

This study employs the OPGD model to identify the core driving factors influencing vegetation NPP. The OPGD model discretizes the data and calculates the explanatory power, represented by the q value, of each driving factor on NPP using the following formula:

$$\mathrm{q}=1-{\displaystyle \frac{{\sum }_{\mathrm{h}=1}^{\mathrm{L}}{\mathrm{N}}_{\mathrm{h}}{\mathsf{\sigma }}_{\mathrm{h}}^2}{{\mathrm{N}\mathsf{\sigma }}^2}}$$

(8)

where q is the explanatory power, with a value range of [0, 1]; N and σ² are the total number of samples and variance, respectively; Nh and σh² are the number and variance of the h-th type of samples, respectively; L is the number of classifications of driving factors. When the q value is closer to 1, it indicates that the driving factor has a stronger explanatory power for NPP.

2.3.5. Residual Analysis

This study calculates three indicators reflecting changes in vegetation NPP: (1) the actual NPP (V_NPP-A) based on MOD17A3HGF data; (2) potential NPP (V_NPP-CC) predicted using a binary linear regression model, random forest regression model, and support vector machine (SVM) regression, representing the predicted vegetation NPP under the influence of CC only; (3) NPP changes caused by HAs (V_NPP-HA), calculated as the difference between potential NPP and actual NPP: NPP(V_NPP-HA) = NPP(V_NPP-CC) - NPP(V_NPP-A). The slopes Sc and Sa of NPP(V_NPP-CC) and NPP(V_NPP-A) are calculated, respectively, through Theil–Sen median trend analysis. These slopes are used to evaluate the impacts of HAs and CC on vegetation NPP and to represent the restoration or degradation status of vegetation. When Sa > 0, it indicates that the actual NPP is increasing; when Sa < 0, the actual NPP is decreasing. Similarly, when Sc > 0, CC leads to an increase in vegetation NPP; when Sc < 0, CC leads to a decrease in vegetation NPP. If Sa - Sc < 0, HAs cause a decrease in vegetation NPP; if Sa - Sc > 0, HAs cause an increase in vegetation NPP. The notation “CC&HA+” indicates that HAs and CC together contribute to an increase in NPP; “HA+” indicates that HAs alone lead to an increase in NPP; “CC+” indicates that CC alone leads to an increase in NPP. Conversely, “CC&HA-” indicates that HAs and CC together result in a decrease in NPP; “HA-” indicates that HAs alone lead to a decrease in NPP; and “CC-” indicates that CC alone leads to a decrease in NPP (

Table 2. Evaluation method for the relative effects of HAs and CC on changes in vegetation NPP.

Change	Kp	Kh	CC Driving	Human Activity Driving	Type
Slope > 0	>	>	${\displaystyle \frac{\left|\Delta V_{\mathrm{NPP}-\mathrm{CC}}\right|\ast 100\%}{\left|\Delta V_{\mathrm{NPP}-\mathrm{CC}}\right|+\left|\Delta V_{\mathrm{NPP}-\mathrm{HA}}\right|}}$	${\displaystyle \frac{\left|\Delta V_{\mathrm{NPP}-\mathrm{HA}}\right|\ast 100\%}{\left|\Delta V_{\mathrm{NPP}-\mathrm{CC}}\right|+\left|\Delta V_{\mathrm{NPP}-\mathrm{HA}}\right|}}$	CC&HA+
	<	>	0	100	HA+
	>	<	100	0	CC+
Slope < 0	<	<	${\displaystyle \frac{\left|\Delta V_{\mathrm{NPP}-\mathrm{CC}}\right|\ast 100\%}{\left|\Delta V_{\mathrm{NPP}-\mathrm{CC}}\right|+\left|\Delta V_{\mathrm{NPP}-\mathrm{HA}}\right|}}$	${\displaystyle \frac{\left|\Delta V_{\mathrm{NPP}-\mathrm{CC}}\right|\ast 100\%}{\left|\Delta V_{\mathrm{NPP}-\mathrm{CC}}\right|+\left|\Delta V_{\mathrm{NPP}-\mathrm{HA}}\right|}}$	CC&HA-
	>	<	0	100	HA-
	<	>	100	0	CC-

).

3. Results

3.1. Spatiotemporal Evolution Characteristics of Vegetation NPP

3.1.1. Temporal Change Characteristics

From 2000 to 2023, the interannual variation in vegetation NPP in the LanXi urban agglomeration exhibited a generally fluctuating upward trend (Figure 3), with an average annual growth rate of 4.26 gC·m⁻²·a⁻¹ and a multi-year mean of 260.23 gC·m⁻²·a⁻¹. The regression fitting curve demonstrated a strong correlation (R² = 0.78, p < 0.01), indicating statistical significance. The lowest NPP value was documented in 2000 at 193.02 gC·m⁻²·a⁻¹, while the highest occurred in 2019 at 321.43 gC·m⁻²·a⁻¹. The most substantial increase in NPP occurred between 2018 and 2019, with a change magnitude of 24.37 gC·m⁻²·a⁻¹, whereas the sharpest decline took place between 2019 and 2020, also with a change magnitude of 24.37 gC·m⁻²·a⁻¹.

Based on the distribution characteristics of vegetation NPP values (Figure 4), the LanXi urban agglomeration experienced three distinct evolutionary stages in the temporal dynamics of NPP: steady improvement, periodic fluctuations, and the expansion of high-value areas. The proportion of low-value areas (NPP < 100 gC·m⁻²·a⁻¹) consistently declined from 20.12% in 2000 to a minimum of 2.67% in 2019. In contrast, high-value areas (NPP > 450 gC·m⁻²·a⁻¹) began increasing rapidly after 2013, reaching 14.68% by 2023. This shift reflects the growing resilience and adaptability of the region’s ecosystem. Concurrently, areas with medium-low NPP values (100–300 gC·m⁻²·a⁻¹) exhibited an overall downward trend, indicating a temporal transition in ecological quality toward medium- to high-value intervals.

3.1.2. Spatial Change Characteristics

In terms of spatial distribution, vegetation NPP in the LanXi urban agglomeration exhibited pronounced heterogeneity throughout the study period, displaying a distinct pattern of “high in the central region and low in the east and west” (Figure 5a). Areas with an annual average NPP exceeding 300 gC·m⁻²·a⁻¹ accounted for 35.2% of the total area and were primarily concentrated in the central, southeastern, and southwestern parts of the agglomeration. In contrast, regions with NPP values below 100 gC·m⁻²·a⁻¹ made up 11.1% of the area, primarily located in the northwestern zones and urban expansion areas.

From 2000 to 2023, the mean NPP across the study area generally increased, with 92% of the region experiencing rising NPP values (Figure 5b). This upward trend was particularly evident in the central and southeastern regions. However, some urban expansion areas within the core agglomeration and portions of the northwest exhibited declining NPP trends, representing 8% of the total area. Meanwhile, the NPP in river and lake regions such as Qinghai Lake remained relatively stable, with no significant changes being observed over the study period.

3.2. Driving Factor Influence Analysis

The annual change slope of vegetation NPP in the LanXi urban agglomeration from 2000 to 2023 ranged from −18.4 to 16.64 (Figure 6a). Areas showing a growing trend in NPP (slope > 0) constituted over 95% of the total area. Regions with moderate increases (slope < 5) were primarily distributed in the western and northeastern parts of the agglomeration, while areas with steeper increases (slope > 5) were predominantly located in the central and southeastern regions, reflecting a general spatial gradient of decreasing NPP growth from southeast to northwest. Areas with declining NPP trends (slope < 0) represented less than 3% of the area and were primarily found in urban expansion zones surrounding the provincial capitals of Lanzhou and Xining, as well as other prefecture-level city centers.

Trend significance analysis revealed that NPP changes were statistically significant across the entire region, with the majority of areas passing the 0.01 significance threshold (Figure 6b). The average Hurst index was 0.649, well above the 0.5 threshold, indicating a strong persistence in the future upward trend of vegetation NPP across the agglomeration (Figure 6c). The Hurst index values ranged from 0.26 to 0.80, with 97.85% of the area displaying indices above 0.5.

By overlaying the spatial distributions of the current NPP change slope and the Hurst index, the study identified four distinct future NPP trend patterns within the LanXi urban agglomeration (Figure 6d): (1) areas with sustained NPP increases accounted for 96.61% of the total area; (2) areas projected to shift from increase to decrease constituted 2.07%, mainly scattered across the central and northwestern parts at the junction of the Qinghai–Tibet and Loess Plateaus; (3) regions with continued NPP decline made up 1.24%, largely concentrated around river systems and urban cores; (4) areas currently exhibiting a decline but expected to reverse into an increasing trend in the future represented only 0.07%.

3.3. Analysis of Driving Factors for Vegetation NPP

3.3.1. Analysis of Driving Factor Influence

The average vegetation NPP from 2000 to 2023 was selected as the dependent variable (Y), while the driving factors (X) included potential evapotranspiration (PET), actual evapotranspiration (EVP), annual precipitation (PRE), digital elevation model (DEM), POP, economic density (GDP), slope (SLOPE), and mean annual temperature (MAT). Drawing on the GD package in R and related research methodologies, each factor was discretized into 3–7 categories [25]. Five discretization methods—the equal interval, quantile, natural breaks, geometric interval, and standard deviation—were tested to identify the optimal classification scheme for each factor. The results revealed significant differences in explanatory power across classification methods. Specifically, PET, PRE, EVP, and GDP were best represented using six categories, whereas DEM, POP, SLOPE, and TEM were optimally classified into seven categories. As a categorical variable, LU did not require further discretization.

The influence of each driving factor X on the dependent variable Y was quantified using the q-statistic, which ranges from 0 to 1. A higher q-value specifies stronger explanatory power of the corresponding factor. As shown in Figure 7, the factors ranked in descending order of explanatory power were as follows: PRE (0.434) > TEM (0.337) > PET (0.327) > LU (0.275) > EVP (0.252) > SLOPE (0.225) > DEM (0.219) > GDP (0.112) > POP (0.097). All factors passed the 0.05 significance threshold. These results indicate that vegetation NPP in the LanXi urban agglomeration, situated within the ecotone between the Qinghai–Tibet Plateau and the Loess Plateau, is jointly affected by climatic–geographical conditions and socioeconomic activities.

The interaction analysis revealed that the spatial heterogeneity of vegetation NPP in the LanXi urban agglomeration results from the coupling effects of multiple driving factors (Figure 8). All individual factors demonstrated enhanced explanatory power following bivariate interactions. The driving factors were categorized into two groups based on their explanatory power [26]: dominant factors (Q ≥ 0.25), including PRE, MAT, PET, EVP, and LU type; important factors (Q < 0.25), including slope, elevation, POP, and GDP. Among the dominant factors, interactions between PRE and MAT, precipitation and LU, and precipitation and PET exhibited strong explanatory power, with q-values reaching 0.6. In contrast, the interaction between GDP and POP yielded the lowest explanatory power, with a q-value of 0.232. Regarding the interaction patterns, dominant factors, such as PRE, MAT, and LU, tended to exhibit nonlinear enhancement when combined with other variables. In comparison, interactions involving important factors generally displayed linear enhancement effects. These results underscore the complex, synergistic influence of both natural and socioeconomic drivers on vegetation NPP across the study area.

3.3.2. Impact of HAs and CC on the Spatiotemporal Evolution of NPP

Based on the results of the Geographical Detector’s factor and interaction detection, as well as previous research on vegetation NPP [27], PRE and MAT were identified as key climatic factors influencing NPP. A CC regression model was constructed using these variables to estimate predicted NPP values. Residuals were then calculated by comparing the predicted and actual NPP values, thereby isolating the effect of HAs by eliminating the effects of climate variability. To evaluate the accuracy of different regression models, the average R² values were compared [28]: binary linear regression achieved an average R² of 0.25, SVM regression achieved 0.45, and random forest regression yielded the highest fit at 0.82 (Figure 9). However, to ensure the robustness of the model, validation was conducted using data from adjacent regions. The resulting decline in the R² value for the random forest model indicated potential overfitting [29]. Considering both model performance and generalizability, the nonlinear SVM regression was ultimately selected for residual analysis to effectively differentiate the relative contributions of HAs and CC to vegetation NPP.

Based on the residual analysis used to quantify the relative contributions of HAs and CC to vegetation NPP changes (Figure 10), the results indicate that HAs had a greater influence than CC in the LanXi urban agglomeration. Specifically, HAs contributed 67.56%, while CC contributed 32.44%.

As shown in Figure 10a, the relative contribution of CC is predominantly positive, with 90% of the total area falling within the 0–50% contribution range. Areas with a contribution rate of 50–100%, mainly located in the southern forested regions at the Gansu–Qinghai provincial border, account for 6.6%. In contrast, negative contributions of CC to NPP are observed in 5.6% of the area, primarily situated in ecological transition zones between the Qinghai–Tibet Plateau and the Loess Plateau, as well as in urbanized zones around provincial capitals.

Figure 10b illustrates that the relative contribution of HAs is largely concentrated in the 50–100% range, covering 92.78% of the study area. A smaller proportion (5.1%) with a positive contribution rate of 0–20% is found mainly in the western and southern ecological transition zones. Negative contributions from HAs, accounting for 4.5% of the area, are primarily distributed in highly urbanized regions near provincial capitals and intensively cultivated zones along central river valleys.

In summary, from 2000 to 2023, both HAs and CC exerted predominantly positive impacts on vegetation NPP in the LanXi urban agglomeration, with HAs serving as the dominant driver.

Based on the driving factor identification framework (Table 1), the drivers of vegetation NPP changes in the LanXi urban agglomeration from 2000 to 2023 were identified, and the results are illustrated in Figure 11. The analysis indicates that vegetation NPP was primarily influenced by the joint effects of HAs and CC, which together accounted for 95.72% of the total area. Within these regions, areas exhibiting an increase in NPP due to the joint influence comprised 94.11%, while areas showing a decrease accounted for 1.61%. These decreasing zones were mainly located at the transitional boundary between the Qinghai–Tibet Plateau and the Loess Plateau. In contrast, areas where NPP changes were driven solely by CC represented 1.03% of the total area, with nearly all of these regions showing positive trends. Regions influenced exclusively by HAs covered 3.35% of the area, of which 0.83% experienced increases in NPP and 2.42% experienced declines. Overall, these findings suggest that over the past 23 years, vegetation NPP dynamics in the LanXi urban agglomeration have been predominantly shaped by the synergistic interactions between HAs and CC.

4. Discussion

4.1. Spatiotemporal Changes in Vegetation NPP in the LanXi Urban Agglomeration

From 2000 to 2023, the average vegetation NPP in the LanXi urban agglomeration was 260.23 gC m⁻² a⁻¹, exhibiting an overall fluctuating upward trend with an average annual increase of 4.26 gC m⁻² a⁻¹. Spatially, NPP displayed a “high in the center, low in the periphery” distribution pattern, consistent with previous findings on the spatiotemporal evolution of vegetation NPP in similar regions [30,31]. Studies conducted in arid ecological transition zones have confirmed the significant spatial heterogeneity of vegetation NPP in such environments [32]. The overall increase in NPP across the LanXi urban agglomeration during this period was largely attributed to the continuous implementation of ecological protection and restoration initiatives, such as China’s large-scale programs for returning farmland to forests, protecting natural forests, and restoring wetlands [33]. These efforts have effectively facilitated vegetation recovery, enhanced ecosystem productivity, and improved the resilience and adaptability of natural ecosystems.

From a temporal perspective, NPP increased steadily from 2000 to 2013, accompanied by a sharp reduction in the proportion of low-value areas, suggesting a marked improvement in regional ecological quality [34]. This positive trend was driven by the joint effects of ecological engineering projects, vegetation restoration, and relatively favorable climatic conditions. However, after 2019, a slight decline in NPP was observed, particularly in urban expansion zones and high-altitude areas in the northwest. This decrease was mainly associated with intensified HAs, LU changes, and climatic fluctuations. Notably, between 2008 and 2010, the NPP growth rate decelerated and even declined locally due to accelerated urbanization and intensive land development in cities such as Lanzhou and Xining. These disturbances disrupted vegetation cover and adversely affected ecosystem productivity. Between 2010 and 2023, NPP rebounded significantly, especially in central and southeastern regions undergoing ecological restoration, indicating a transition of ecosystem quality toward medium- to high-value intervals.

For NPP estimation, prior research has validated remote-sensing-based NPP data using field measurements, such as grassland quadrat biomass and carbon cycling reference datasets from typical Chinese forest ecosystems [35], thereby enhancing the accuracy and reliability of findings.

In terms of spatial distribution, higher NPP levels were found in the southeastern and southwestern parts of the LanXi urban agglomeration, primarily due to favorable natural conditions, denser vegetation cover, and effective implementation of ecological restoration projects. In contrast, lower NPP values occurred in high-altitude northwestern regions and urban expansion zones, where ecosystem productivity was constrained by both geographical limitations and anthropogenic disturbances [7]. These patterns highlight that while ecological protection policies have substantially improved regional ecosystem quality and boosted vegetation productivity, there remain risks of NPP decline in urbanized and ecologically fragile areas. Continued ecological conservation and sustainable land management are essential to safeguard long-term ecosystem health and resilience.

4.2. Analysis of Vegetation NPP Change Trends in Ecological Transition Zones

This study reveals the vegetation NPP trends in the LanXi urban agglomeration from 2000 to 2023, highlighting a significant overall upward trend, with over 95% of the area showing NPP increases. While consistent with some prior research [36], these results exhibit unique spatial heterogeneity and temporal dynamics attributable to the LanXi region’s distinct geographical setting as a junction between the Qinghai–Tibet Plateau and the Loess Plateau.

Liu Hui et al. [37] reported a slight decline in vegetation NPP in plateau–basin junctions due to severe desertification driven by CC; however, our findings indicate a significant NPP increase in the Qinghai–Tibet–Loess Plateau transition zone. Unlike studies of other urban agglomerations [38], which found that intensified HAs generally reduce vegetation NPP, this study shows that human activity areas do not fully overlap with low-NPP zones, emphasizing the critical influence of natural geographical factors on NPP distribution. Compared to the Qilian Mountain region [39], the LanXi urban agglomeration demonstrates a higher proportion of NPP-increasing areas, especially in the southeastern and central zones, reflecting stronger ecosystem recovery. Notably, areas with a Hurst index above 0.5 account for 97.85%, indicating persistent future growth in vegetation productivity. This positive trend aligns closely with ongoing regional ecological protection policies and favorable natural conditions, underscoring significant advances in ecosystem restoration within the urban agglomeration.

As an ecological transition zone, LanXi exhibits spatially uneven vegetation NPP trends, illustrating both vulnerability and resilience within its ecosystems. The southeastern region responds well to climate and policy drivers with marked recovery, whereas northwestern high-altitude and urban expansion areas show lower productivity and slower ecological restoration, necessitating enhanced long-term ecological monitoring and management. Future research should employ high-resolution remote sensing and multifactor coupling models to more precisely analyze spatiotemporal dynamics and driving mechanisms of vegetation NPP, particularly in areas experiencing decline. Such efforts will provide essential scientific guidance for the sustainable management and development of regional ecosystems.

4.3. Comparison of Linear and Nonlinear Residual Analysis

Comparing linear and nonlinear regression models allows the identification of the best fit for vegetation NPP simulation. In this study, R² was used to evaluate model performance, revealing that nonlinear regression consistently outperforms linear regression, exhibiting higher fitting accuracy in regional NPP modeling, in line with previous research [40]. The LanXi urban agglomeration faces dual challenges of urban development and ecological protection of high-altitude and fragile northwest zones, where HAs and CC are strongly coupled. Linear regression models show significantly lower fitting accuracy, with median R² values as low as 0.2, failing to capture the complex climate–NPP relationship in ecological transition zones.

Notably, natural conditions and human activity intensity vary widely across regions, leading to uneven data distribution. In urban expansion areas with intense human activity, some nonlinear residual regression models suffer from overfitting. Excessively high R² values here may obscure the true influence of HAs on vegetation NPP. Conversely, in the northwest’s high-altitude and ecologically fragile zones, where human disturbance is minimal and climate factors dominate, nonlinear models perform well by effectively capturing complex nonlinear climate–NPP responses, whereas linear models yield lower R² and poorly reflect these dynamics.

In heavily urbanized areas such as Lanzhou and Xining, LU changes driven by HAs primarily impact vegetation NPP [41]. Nonlinear models achieve significantly higher R² values than linear ones, sometimes approaching or exceeding 0.4, indicating strong fitting performance. However, this high accuracy may come with overfitting risks, as complex nonlinear models fitting numerous variables can misinterpret noise or random fluctuations, undermining prediction stability.

By comparing multiple models, an appropriate residual analysis approach can be selected that balances statistical rigor and theoretical relevance for capturing vegetation NPP dynamics in complex ecological transition zones.

4.4. Impact of HAs and CC on NPP in Ecological Transition Zones

The optimal parameter results from Geographical Detector factor detection indicate that vegetation NPP is directly or indirectly influenced by PRE, PET, or EVP, MAT, and LU, which is consistent with studies in China [42] and the American mountain ranges [43]. Overall, natural factors exert a stronger impact on vegetation NPP than HAs. Rainfall and temperature closely correlate with vegetation growth and health [44], directly affecting NPP levels. However, residual analysis reveals that the influence of HAs has increased over time, surpassing that of natural factors, aligning with findings from Guizhou Province, China [45].

Interaction analysis shows that combinations of PRE with average temperature, LU, and again average temperature significantly enhance explanatory power for NPP, with q-values reaching 0.6. Conversely, the interaction between GDP and POP has the least explanatory power. These results highlight the importance of nonlinear coupling between climate factors and LU in driving vegetation NPP changes in the LanXi urban agglomeration, validating the selection of precipitation and temperature as key CC factors [46]. The nonlinear enhancement effect, particularly between precipitation and temperature, underscores how sufficient rainfall combined with suitable temperatures significantly promotes vegetation growth and increases NPP [47].

Through residual analysis, this study quantitatively assessed the relative contributions of HAs and CC to vegetation NPP changes, finding that HAs’ impact significantly exceeds that of CC, which is consistent with related research [48]. Unlike Geographical Detector results that capture spatial differentiation patterns, this study’s contribution rates incorporate future development trends, reflecting the overall trajectory of HAs. The LanXi urban agglomeration, situated at the junction of the Qinghai–Tibet Plateau and Loess Plateau, is influenced by both HAs and CC, with varying degrees across geographic regions, a conclusion supported by research in northern China [49].

Specifically, the LanXi urban agglomeration has not experienced a decline in vegetation NPP due to rapid urbanization, contrasting with the findings of Liu et al. [50]. Instead, ecological restoration projects driven by HAs [51] have significantly enhanced vegetation coverage and productivity, resulting in local impacts of HAs on vegetation NPP in ecological transition zones that exceed those of CC. HAs and CC exhibit synergistic effects: in heavily forested areas, increased precipitation translates more efficiently into vegetation growth, while in agricultural lands, precipitation changes more directly influence crop yields and land management practices, thereby indirectly affecting vegetation NPP. The relative impacts of HAs and CC on vegetation NPP vary markedly across regions, reflecting the distinct characteristics of the ecologically fragile junction between the Qinghai-Tibet Plateau and the Loess Plateau.

Although this study focuses on the LanXi urban agglomeration, a representative ecological transition zone in Northwest China, the methodologies and key findings have broad relevance to similar arid and semi-arid ecological transition zones and rapidly urbanizing regions worldwide. Urban agglomerations in Central Asia [52], Northern Africa [53], the southwestern United States, and the Mediterranean similarly face complex ecosystem challenges under the combined pressures of HAs and CC [54]. The integrated nonlinear residual model and optimized geographical detector effectively clarify the multi-factor coupling mechanisms driving vegetation productivity across different spatial scales and locations, providing valuable methodological insights for cross-regional and interdisciplinary ecosystem management and sustainable land-use planning. These findings offer scientific evidence and decision-making support for urban agglomerations in global arid and ecological transition zones to address CC, evaluate human impacts, and design effective ecological restoration and conservation strategies

4.5. Limitations and Future Prospects

Although this study provides a relatively comprehensive analysis of the spatiotemporal changes, driving factors, and future trends of vegetation NPP in the LanXi urban agglomeration, several limitations and uncertainties remain. Future research should address the following aspects:

(1)

Limited temporal scope: While data from 2000 to 2023 reveal discernible trends, this period is insufficient to fully capture the long-term variation patterns of vegetation NPP in the LanXi urban agglomeration. To improve future trend predictions, subsequent studies should incorporate longer time-series datasets, advanced modeling techniques, and scenario-based projections under various CC conditions.

(2)

Ecosystem complexity: This study primarily focuses on NPP changes, yet ecosystem dynamics are multi-factorial and multi-layered. Future research should explore the comprehensive impacts of additional ecological factors, such as soil types, water resource distribution, and others, on vegetation NPP, especially within complex ecological transition zones. It is also important to note that MODIS NPP data products inherently contain some uncertainties. Due to limitations, this study was unable to fully validate NPP results with field measurements, as performed in other studies. Enhancing validation with ground-truth data would improve the rigor of future work.

(3)

Refined spatial analysis: Although this study examined the LanXi urban agglomeration as a whole, it lacks detailed quantitative analysis of NPP changes in specific local areas, such as urban expansion zones and the Qinghai–Tibet Plateau–Loess Plateau junction. Future investigations should integrate remote sensing technologies with ground observations to conduct more detailed analyses of NPP variation across various LUs and climatic conditions, further elucidating the interactions between HAs and natural factors.

(4)

Integration of multiple models: Future studies should incorporate a broader range of climate and socioeconomic models to enable multi-scenario forecasting. The application of machine learning and other advanced computational techniques can enhance the precision of NPP change predictions. Additionally, improving the modeling of interaction effects among NPP drivers will further increase the accuracy and reliability of future predictions.

5. Conclusions

This study applied trend analysis, stability analysis, OPGD, and residual analysis to examine the spatiotemporal variations and driving factors of vegetation NPP in the LanXi urban agglomeration, an ecological transition zone, from 2000 to 2023. The main conclusions are the following:

(1)

From 2000 to 2023, vegetation NPP showed an overall fluctuating upward trend, with an average annual increase of 4.26 g C m⁻² a⁻¹ and a multi-year mean of 260.23 g C m⁻² a⁻¹. Spatially, NPP exhibited significant heterogeneity, with higher values being concentrated in the central region and lower values in the east and west. The proportion of low-NPP areas declined notably, while high-NPP areas increased significantly.

(2)

Temporally, about 95% of the region experienced increasing NPP, especially in the central and southeastern zones, whereas slight declines appeared in the northwest and urban expansion areas. The upward trend demonstrated strong persistence.

(3)

OPGD analysis identified PRE, MAT, PET, and LU as key drivers explaining vegetation NPP variability. Their interactions further enhanced the explanatory power.

(4)

Comparing linear and nonlinear residual analyses revealed the nonlinear model’s clear superiority. Vegetation NPP was mainly governed by the combined influence of HAs and CC, jointly affecting 95.72% of the area. Sole influences of HAs and CC accounted for 1.03% and 3.35%, respectively.