Data processing using GEE.

Access the Google Earth Engine (GEE) platform via its website (https://earthengine.google.com/) and create a new script in the code editor [31,32,33]. Input the code necessary for downloading and processing data related to vegetation NPP, evapotranspiration, total solar radiation, and population density. After executing the script, download the datasets for vegetation NPP, evapotranspiration, total solar radiation, and population density over the Tibetan Plateau for the period from 2001 to 2021, all at a resolution of 500 meters. Subsequently, calculate the trends in vegetation NPP and conduct trend tests. The results will be saved as GeoTIFF files and exported to a designated folder in Google Drive [33]. Comprehensive details regarding the calculation procedures can be found in the supplementary materials.

Coefficient of variation (Cv).

The coefficient of variation (Cv), also referred to as the coefficient of dispersion, is a key metric used to evaluate the stability and variability of data over time or across space [34,35]. As a dimensionless measure, Cv facilitates comparisons without needing to reference the mean value of the dataset. Consequently, when assessing the stability of vegetation NPP, the ratio of the standard deviation to the mean serves as an effective comparison [36]. The formula for calculating Cv is as follows:

(1)

where, Cv is the coefficient of variation for vegetation NPP, σ is the standard deviation of vegetation NPP, representing the degree of data dispersion, and u is the mean NDVI, representing the average vegetation NPP over a certain period or space. A higher Cv indicates greater variability and less stability in vegetation NPP, whereas a lower Cv suggests that vegetation NPP is relatively stable.

Hurst index.

The Hurst index, a statistical measure of long-term dependence in time series data, is applied in this study to evaluate the persistence or anti-persistence of vegetation NPP trends over time. Following the method proposed by Hurst [37] and refined in subsequent studies,we compute the Hurst index using the NPP time series data [38,39,40]. The process involves calculating the cumulative deviation series, range, and standard deviation, followed by estimating the ratio R(T)/S(T)≌R/S and fitting it to the scaling relationship R/S∝T^H, the calculation of the Hurst index are as follows:

(1) Given Time Series.

Let the time series be {NPP(t)}, where t = 1, 2, …, n.

(2) Compute the mean.

The mean of the time series is calculated as:

(2)

(3) Construct the cumulative deviation series

The cumulative deviation series X_(x,T) represents the cumulative deviation of the series from its mean:

(3)

(4) Calculate the range series:

The range R_(T) is the difference between the maximum and minimum of the cumulative deviation series:

(4)

(5) Calculate the standard deviation series:

The standard deviation S_(T) of the original time series is computed as:

(5)

(6) Calculate the Hurst index:

The Hurst index H is estimated using the following relationship:

(6)

This allows us to classify the time series into three categories: anti-persistent (0<H<0.5), random (H=0.5), and persistent (0.5<H<1). Through this analysis, the study quantifies the degree of persistence in NPP fluctuations, offering critical insights into how vegetation dynamics respond to environmental and anthropogenic influences, particularly under the context of climate change [41,42].

Trend analysis.

The Sen’s slope estimator and Mann-Kendall trend test are non-parametric statistical methods extensively applied to analyze trends in time series data, first introduced by Hirsch and Slack [43]. In recent years, studies have increasingly combined these techniques with other methods to enhance the precision of trend assessments. For instance, research by De Jong et al.[44]and Gocic and Trajkovic highlights the effectiveness of integrating these approaches in uncovering trends within time series datasets [45]. Owing to their complementary strengths, these methods have gained broad applications in remote sensing data analysis. In this study, these methods are employed to detect the spatiotemporal variations in NPP across the research area, providing robust trend estimates for long-term changes in vegetation productivity.

Sen’s slope estimation. Sen’s slope estimation is a reliable non-parametric approach for identifying trends in time series datasets [45,46]. In this study, the method is applied by first arranging observed NPP values in chronological order, followed by dividing the data into several groups. The slopes between all data points within each group are calculated, and the median slope is selected as the group’s trend estimate. The overall NPP trend is determined by calculating the median slope across all groups [47]. This approach reduces the influence of outliers and extreme values, which is critical for ensuring accurate trend detection in large-scale NPP time series datasets [48].

The formula for calculating Sen’s trend estimation is as follows:

(7)

where, β represents the trend in vegetation NPP, and i and j refer to the indices of the time series. NPP_i and NPP_j are the values of vegetation NPP at time series i and j, respectively. When β>0, it indicates an increasing trend in vegetation NPP, whereas β<0, reflects a decreasing trend. The larger the absolute value of the slope, the more significant the change in vegetation NPP [45,47].

Mann-Kendall trend test. The Mann-Kendall trend test is a commonly used non-parametric method for evaluating the significance of trends in time series data [49,50,51,52]. This method has been extensively applied in ecosystem NPP studies, particularly for assessing long-term impacts of climate change and land-use changes on NPP [53,54]. In this study, the Mann-Kendall trend test is used to evaluate the significance of spatiotemporal changes in vegetation NPP, enabling a comprehensive understanding of trend dynamics and their underlying drivers.

The calculation formula for the Mann-Kendall trend test is as follows:

(8)

(9)

(10)

(11)

Where, NPP_i and NPP_j represent the NPP values for a pixel in year i and year j, respectively, with n denoting the length of the time series, which spans 21 years. The standardized normal statistic is represented by Z, and S is the trend statistic, which approximately follows a normal distribution, calculated using NPP values for the period from 2001 to 2021. The function sng is the sign function, and Var(S) refers to the variance of the statistic. This study uses a two-tailed test with a significance level α=0.05, which yields a critical value of Z₁−α/2= ±1.96. When the absolute value of the Z-statistic exceeds 1.65, 1.96, or 2.58, the trend is significant at the 90%, 95%, and 99% confidence levels, respectively [14,15].

OPGD.

The Optimal Parameters-based Geographical Detector (OPGD) model is a statistical method designed to assess spatial heterogeneity and its driving factors. By optimizing parameter selection, the OPGD model improves upon the traditional Geographical Detector (GD) model, allowing for more accurate quantification of the influence of both natural and human activities on spatial patterns.

Parameter optimization. The OPGD model utilizes optimization algorithms that adjust the detector’s parameters automatically according to the characteristics of each variable, ensuring the model’s suitability for complex spatial analysis [55]. For factor detection, this study employs an optimized discretization approach, calculating the Q value through combinations of breakpoints for continuous variables. The factor detector assesses different classification methods and breakpoint combinations, selecting the one that maximizes the Q value as the optimal parameter set [55]. This method enhances the identification of spatial stratified heterogeneity and highlights the influence of key continuous variables on spatial variation [55,56,57]. By choosing the parameter set with the highest Q value, the model accurately detects spatial heterogeneity in vegetation NPP on the Tibetan Plateau and identifies the factors driving these changes [58]. Optimizing the discretization parameters increases the explanatory power of continuous geographical variables, offering deeper insights into the spatial distribution and dynamics of NPP [55].

Spatial differentiation analysis. The OPGD model assesses the importance of each variable by comparing the variance between the observed values in the study area and the variable layers under different parameter settings[55,59]. It calculates the Q value to determine the explanatory power of each factor in relation to spatial heterogeneity, identifying the primary drivers[55]. The process involves the following steps: First, the vegetation NPP layers are overlaid with variable factor layers to uncover their spatial relationships. This helps in understanding the association between vegetation NPP and various factors. Second, the spatial classification of different factors is performed by grouping them into regions or categories, providing a clearer view of the factors’ impact across different spatial scales. Finally, a significance test is conducted to compare the mean differences between various factors, which helps determine their contribution to changes in vegetation NPP and identify the most influential factors driving NPP variations.

The explanatory power (Q) of the factors is calculated using the following model:

(12)

where, Q represents the explanatory power of the factor on vegetation NPP, with a range of [0, 1]; the larger the value, the stronger the factor’s explanatory power for vegetation NPP. h = 1,..., L denotes the layers of vegetation NPP and influencing factors. N_h is the number of units in layer h. N is the total number of units in the region. and σ² are the variances of vegetation NPP in layer h and the entire region, respectively. SSW represents the sum of the within-layer variances. SST is the total variance across the entire region.

The variance of the regional Y value is calculated as follows:

(13)

where, Y_j and represents the value of sample j and is the mean value of Y for the entire region.

Multifactor interaction analysis. The OPGD model allows for the analysis of interactions between natural and human factors, shedding light on their combined effects on spatial phenomena. By exploring these interactions, the model helps to clarify the complex relationships between various driving forces and identifies the key factors influencing NPP. The factor detector calculates the relative importance of each variable and reveals how overlapping spatial variables interact. Interaction effects emerge when two variables are spatially combined, reflecting their collective impact [55].

The interaction detector compares the Q value of the combined variables to the individual Q values, determining whether the interaction is enhancing, weakening, or independent. Five interaction types are identified: nonlinear weakening, single-variable weakening, double-variable enhancement, independence, and nonlinear enhancement [55,59].

Risk detection. Risk detection analyzes how a specific factor (such as environmental or socio-economic variables) influences the distribution of a spatial phenomenon (e.g., disease, pollution, natural disasters) [55,59]. The method compares the factor’s distribution across regions with the target phenomenon, determining whether the factor significantly affects spatial heterogeneity. The t-statistic is used to test whether the mean NPP in subregion h differs significantly from the overall mean:

(14)

where, represents the mean NPP attribute value within subregion h, n_h is the number of samples within subregion h,Var denotes the variance of the NPP values. This formula assesses whether the mean NPP in subregion h significantly differs from the overall region’s mean.

Ecological detection. Ecological detection compares the influence of different ecological factors on the spatial distribution of NPP, determining whether certain variables have a more pronounced effect [55,59]. The F-statistic is employed to evaluate whether significant differences exist between the impacts of these factors on NPP distribution:

(15)

(16)

(17)

where, N_x1 and N_x2 are the sample sizes of two factors, and SSW_x1and SSW_x2 represent the sums of intra-layer variances. L₁ and L₂ indicate the number of stratifications for variables x₁ and x₂, respectively.

Spatial distribution.

The spatial distribution of vegetation NPP across the Qinghai-Tibet Plateau reveals pronounced regional disparities, varying significantly between different areas and over time. Higher NPP values are predominantly observed in the southeastern regions of the plateau, while lower values are concentrated in the northwest. Favorable climatic conditions in the eastern and southern regions, including increased humidity and denser vegetation, contribute to relatively high NPP levels. In contrast, the drier climates and higher altitudes in the western and northern areas result in significantly lower NPP, with some regions displaying values close to zero (Fig 3).

Download:

• PPT
  PowerPoint slide
• PNG
  larger image
• TIFF
  original image

Fig 3. Vegetation NPP spatial pattern changes in the Qinghai-Tibet Plateau from 2001 to 2021.

https://doi.org/10.1371/journal.pone.0320370.g003

Temporal trends in NPP distribution.

As shown in Fig 3, the spatial patterns of NPP exhibit notable changes over the years: In 2001, the maximum NPP value was 1905 gC m^-2·a^-1, primarily concentrated in the southeastern portion of the plateau, reflecting robust vegetation growth in this region (Fig 3A). By 2011, the maximum NPP had increased to 1947 gC m^-2·a^-1, indicating improved vegetation productivity, particularly in the southeast [20,60] (Fig 3B). This increase is likely associated with favorable climatic changes, such as higher precipitation and warmer temperatures, which fostered an expansion of vegetated areas [9,60].However, by 2021, the peak NPP value had declined to 1731 gC m^-2·a^-1, suggesting reduced productivity over the preceding decade. This decline may be attributed to adverse factors such as climate change, leading to droughts and extreme weather events, as well as land use changes like overgrazing and urban development, which have likely suppressed vegetation growth (Fig 3C). From 2001 to 2021, the highest NPP values consistently occurred in the southeastern part of the plateau, particularly near the Hengduan Mountains and the Sichuan Basin. Over the study period, the maximum recorded NPP in this region reached 1810 gC m^-2·a^-1, with an average of 724.72 gC m^-2·a^-1 (Fig 3D).

Regional distribution.

The central and northwestern regions exhibited significantly lower NPP values, averaging 34.21 gC m^-2·a^-1, particularly in the arid and alpine zones of the northwest. Southeastern lowlands supported abundant vegetation and high productivity, while the sparse or absent vegetation in the high-altitude interior and northern regions resulted in much lower NPP levels.

In summary, the distribution of NPP on the Qinghai-Tibet Plateau exhibits a distinct zonal pattern [20]. The southeastern fringes maintain higher productivity due to favorable climatic and environmental conditions, whereas the interior regions, characterized by harsh climates and sparse vegetation, display substantially lower NPP values (Fig. 3D).

Spatial fluctuation.

The spatial variability of vegetation NPP on the Qinghai-Tibet Plateau, expressed as the coefficient of variation (Cv), ranges from 0 to 4.58 gC m^-2.a^-1 (Fig 5A). Based on the classification schemes proposed in previous studies (Peng et al., 2014; [5,6]), Cv is divided into five fluctuation levels: low (0–0.05 gC m^-2.a^-1), relatively low (0.05–0.13 gC m^-2.a^-1), moderate (0.13–0.23 gC m^-2.a^-1), relatively high (0.23–1.03 gCm^-2.a^-1), and high (1.03–4.58 gC m^-2.a^-1) (Fig. 5A).

Download:

• PPT
  PowerPoint slide
• PNG
  larger image
• TIFF
  original image

Fig 5. Cv (A), Hurst index (B), Sen (C), and significance test (D) of vegetation NPP on the Qinghai-Tibet Plateau from 2001 to 2021.

In D, the scale from ‐3 to 3 represents the classification of the NPP trend significance test, corresponding to the categories outlined in Table 3.

https://doi.org/10.1371/journal.pone.0320370.g005

Download:

• PPT
  PowerPoint slide
• PNG
  larger image
• TIFF
  original image

Table 3. Statistical summary of vegetation NPP trend from 2001 to 2021.

https://doi.org/10.1371/journal.pone.0320370.t003

Spatially, low and relatively low fluctuation areas dominate the plateau, covering about 73.78% of its total area. These zones are primarily located in the arid northern desert regions. Despite their sparse vegetation, these regions exhibit minimal NPP variability, likely due to the inherent resilience of the vegetation to harsh environmental conditions.

Moderate fluctuation zones, accounting for 23.77% of the plateau’s area, are more widely distributed, occurring across the northeast, central, southeast, and southwest regions. These areas show increased NPP variability, which may result from climatic fluctuations and anthropogenic influences.

High and relatively high fluctuation zones are limited, comprising only 2.45% of the plateau’s area. These regions are predominantly situated in the southeast, where factors such as complex topography and intensified human activities contribute to the higher variability in NPP.

In summary, the Qinghai-Tibet Plateau displays notable spatial variability in vegetation NPP. Most of the plateau exhibits stable NPP dynamics, while only a small fraction experiences significant fluctuations due to environmental and human-induced factors.

Change trend analysis.

From 2001 to 2021, the Hurst index of vegetation NPP on the Qinghai-Tibet Plateau ranged from ‐0.08 to 0.96, with a mean value of 0.41 and a standard deviation of 0.08. To interpret these values, the index is divided into five categories: <0 (‐0.08 to 0), 0–0.3, 0.3–0.5, 0.5–0.7, and >0.7 (0.7–0.96) (Fig. 4B).

The spatial distribution of the Hurst index reveals chaotic patterns with no evident clustering, reflecting the heterogeneity of vegetation dynamics across the plateau. Regions where H<0.5 account for 40.63% of the area, suggesting that the NPP time series in these zones are anti-persistent, meaning that future trends are likely to reverse previous patterns. In contrast, areas with H >0.5 comprise 8.80%, indicating persistence in NPP trends, where future changes are expected to follow historical trajectories. The remaining 50.57% of the plateau exhibits H =0.5, signifying random NPP dynamics with no clear tendency for persistence or anti-persistence, making future trends difficult to predict.

These findings highlight the complexity and variability of vegetation changes across the Qinghai-Tibet Plateau. The significant proportion of anti-persistent and random regions suggests that future NPP trends may be highly dynamic, with potential reversals and uncertainties dominating the plateau’s vegetation development.

Sen trend and MK test.

The fitted slope of vegetation NPP on the Qinghai-Tibet Plateau ranges from −423.93 to 299.36 gC m^-2.a^-1, with 62.93% of the area showing a positive slope, indicating an overall increasing trend in NPP (Fig 5C). Significant increases or stability in NPP are observed in the northeastern, central, eastern, and southeastern regions of the Qinghai-Tibet Plateau, whereas declines are seen in the southwestern, southern, and parts of the eastern regions.

According to Fig 5D and Table 3, most areas of the Qinghai-Tibet Plateau show stable or slightly increasing NPP trends, with significantly decreasing regions being relatively scarce. While NPP has decreased in some areas, the overall trend indicates ecosystem stability or improvement, particularly with prominent increases in the northern and eastern regions. Areas with significant increases in vegetation NPP (fitted slope greater than 2.54 gC m^-2.a^-1, Z-value greater than 1.96, p<0.01 or 0.01≤p<0.05) are marked in dark green (β = 2), covering approximately 707,963 km², or 27.85% of the total area, mainly concentrated in the northern, western, and southern margins. This may reflect the impact of ecological restoration measures and the extended growing season due to climate warming.

Areas of sharp NPP decline (fitted slope less than −4.48 gC m^-2.a^-1, Z-value less than −1.96, p<0.01 or 0.01≤ p <0.05) are marked in red (β = −2), primarily located in the eastern and southern parts of the plateau (Table 3, Fig 3D), where human activity is frequent. The decline in NPP in these areas could be related to human exploitation, climate change, and vegetation degradation.

Regions with no significant change in NPP (fitted slope between −2.99 and 0.68 gC m^-2.a^-1, with −1.65 ≤ Z ≤ 1.65, p ≥ 0.05) are shown in yellow. These areas cover the largest portion, with an area of 784,630 km², accounting for 30.86% of the total, mainly distributed in the central and northern parts of the plateau (Table 3, Fig 5D), likely due to the limited number of NPP pixels and insignificant changes.

Slightly increasing NPP areas (fitted slope between 0.69 and 2.54 gC m^-2.a^-1, 1.65< Z <1.96, and p≥0.05) and slightly decreasing areas (fitted slope between −4.48 and −2.99 gC m^-2.a^-1, −1.65< Z <‐1.96, p≥0.05) are mainly concentrated in parts of the central and northern plateau and parts of the eastern and southern plateau, respectively (Table 3, Fig 5D). These regions are marked in orange (β = −1) and light green (β = 1), accounting for 51.55% and 1.32% of the total area, respectively, which may be associated with slight climate fluctuations or minimal human impacts.

Variables discretization.

The OPGD model in this study was applied to analyze the spatial explanatory variables associated with vegetation NPP changes. The model is capable of integrating both categorical and continuous variables,where categorical variables are incorporated directly into the geographical detector model. For continuous variables, discretization was performed through Q optimization, applying different grading methods to identify breakpoints and determine the best parameter combinations. The results reveal that the optimal parameters and number of breakpoints differ across various explanatory factors (Fig 6A, B).

Download:

• PPT
  PowerPoint slide
• PNG
  larger image
• TIFF
  original image

Fig 6. Exploration of explanatory variables for vegetation NPP Change based on OPGD

: the process of spatial data discretization parameter optimization (A) and the results (B). The meanings of X1-X20 are provided in Table 1.

https://doi.org/10.1371/journal.pone.0320370.g006

Variables such as X1, X2, X7, X9, and X13 were classified into nine categories using the natural breaks method. X3 and X8, however, were divided into nine levels via the equal interval method. Meanwhile, variables like X4, X6, X14, X18, X19, and X20 were split using the standard deviation method, also into nine levels. Variables X10, X11, X12, and X16 were grouped according to the quantile method, again with nine levels. Lastly, X15 was categorized into eight levels and X17 into six levels, both based on the quantile method.

Influence of detection factors.

The findings from the factor analysis reveal that the variables elevation, total solar radiation, annual average temperature, and accumulated temperature above 0 °C exhibit the highest Q values of 0.5696, 0.5634, 0.5367, and 0.5327, respectively. Each of these factors accounts for over 53% of the overall variation in NPP, signifying their substantial influence (Fig 7A). This indicates that natural elements like elevation, solar radiation, and temperature are key contributors to changes in NPP, highlighting their dominant role.

Download:

• PPT
  PowerPoint slide
• PNG
  larger image
• TIFF
  original image

Fig 7. Calculation results of the OPGD model.

(A) Q values of individual variables; (B) correlation coefficient matrix of Q value; (C) interaction detection results; (D) ecological detection outcomes.

https://doi.org/10.1371/journal.pone.0320370.g007

Regarding human activity factors, forest distance and road distance show Q values of 0.4139 and 0.3254, respectively, both surpassing 32% in explanatory power, suggesting that these distances significantly impact NPP variability. Additionally, population density and GDP density demonstrate Q values of 0.2839 and 0.2740, also exceeding 27%, indicating a meaningful effect. The Q values for annual mean precipitation and land use intensity are recorded at 0.2637 and 0.2462, respectively, suggesting that both natural and socio-economic factors contribute to NPP variation.

In contrast, the Q values for slope, aspect, soil moisture, evapotranspiration, distance to water bodies, and proximity to built-up areas are relatively low, measuring 0.1683, 0.0017, 0.1084, 0.0372, 0.1053, and 0.0925, respectively. These values indicate an explanatory power below 17%, suggesting that these factors have a minimal individual influence on NPP (Fig 7A).

In conclusion, key natural factors such as elevation, solar radiation, and annual mean temperature are essential to understanding NPP variations. Simultaneously, human activity factors including forest distance, road distance, population density, and GDP density play significant roles, while the impacts of slope, aspect, soil moisture, and evapotranspiration are comparatively limited.

Interaction detection of factors.

This analysis employs interaction detection to assess how various natural and anthropogenic factors collectively influence NPP, determining whether they enhance, diminish, or act independently on this dependent variable (Fig 7B, C).

The size of the circles indicates the strength of the interactions, with red representing nonlinear enhancement and yellow representing bilinear enhancement in Fig 7C. The findings from Fig 6B and C reveal significant interactions among elevation, annual precipitation, annual temperature, and solar radiation.Notably, the interaction between elevation and annual precipitation is the most pronounced, indicating that precipitation at high altitudes significantly affects NPP. This combined effect exceeds the influence of each individual factor, highlighting the critical role of water and heat availability in vegetation growth at elevated locations. For instance, the interaction strengths rank as follows: X1∩X6 (0.7373) > X1∩X5 (0.6783) > X1∩X11 (0.6458).

In addition, the interplay between population density and GDP, along with land use intensity, shows significant effects, particularly the interaction between population density and GDP. This finding suggests that in economically developed and densely populated regions, human activities substantially impact NPP, especially when land use changes occur, amplifying the overall influence of human activities. The interaction strengths for this pairing are as follows: X12∩X13 (0.6116) > X12∩X14 (0.5332) (Fig 7B, C). Moreover, the relationship between road distance and both forest distance and cropland distance exhibits enhancement effects, particularly highlighted by the interaction of road distance and forest distance. This relationship emphasizes how road construction’s spatial dynamics with forest areas critically influence NPP, likely due to deforestation impacts from road development. The interaction strengths are illustrated as X15∩X17 (0.5284) > X15∩X16 (0.5107). Furthermore, the interaction of annual temperature with land use intensity is notably robust, slightly surpassing that between annual temperature and population density. This suggests that in regions with varying land use intensities, temperature significantly impacts vegetation productivity, especially in warmer areas where land use changes exacerbate NPP influences, demonstrated by X5∩X14 (0.5103) > X5∩X12 (0.4948) (Fig 7B, C).

Overall, there exists a marked interactive enhancement effect between natural and anthropogenic factors affecting NPP, characterized by mutual and nonlinear reinforcement. It is evident that no natural factor operates in isolation from others (Fig 7B, C). The interactions among natural factors, such as elevation, precipitation, temperature, and solar radiation, are particularly strong, and the synergy between human activity factors, like population density, GDP density, and land use intensity should also be recognized.

Additionally, ecological detection serves to evaluate the relative significance of natural factors impacting NPP. The results from this detection, shown in Fig 7D, indicate whether statistically significant differences exist among pairs of factors. A marked difference is represented by “Y,” while the absence of such a difference is indicated as “N” (Fig 7D). The analysis reveals significant differences among all factors concerning NPP spatial distribution, except between X7 and X5. These notable variations highlight the necessity of considering multiple factors comprehensively to understand the spatial variations in NPP effectively.

Detection of factor indicative roles.

The risk detector analysed the performance of influencing factors in environments suitable for changes in vegetation NPP and conducted statistical significance tests at a 95% confidence level (Fig 8). Higher NPP values indicate that natural and human activity factors are more favourable to NPP variation. The results show that there are significant differences in the impact of different intervals of explanatory variables on vegetation NPP variation (Fig 8).

Download:

• PPT
  PowerPoint slide
• PNG
  larger image
• TIFF
  original image

Fig 8. Risk detection (vegetation NPP variation is classified into three levels: high (red), medium (grey), and low (blue), based on the identified variables).

https://doi.org/10.1371/journal.pone.0320370.g008

Elevation, total solar radiation, distance to roads, distance to cropland, distance to forestland, river network density, and distance to built-up areas all reach their highest NPP values in the first classification interval, with respective values of 812.65 gC m^-2.a^-1, 714.64 gC m^-2.a^-1, 264.89 gC m^-2.a^-1, 399.69 gC m^-2.a^-1, 47270491.22 gC m^-2.a^-1, and 330.14 gC m^-2.a^-1. In contrast, slope, soil moisture, annual mean temperature, annual precipitation, 0°C accumulated temperature, population density, GDP density, land use intensity, and distance to water bodies reach their highest NPP values in the ninth classification interval, with values of 434.19 gC m^-2.a^-1, 625.19 gC m^-2.a^-1, 860.20 gC m^-2.a^-1, 456.42 gC m^-2.a^-1, 882.39 gC m^-2.a^-1, 429.22 gC m^-2.a^-1, 412.99 gC m^-2.a^-1, 419.71 gC m^-2.a^-1, and 353.06 gC m^-2.a^-1, respectively. Evapotranspiration and distance to grassland reach their highest NPP values in the sixth classification interval, with values of 257.41 gC m^-2.a^-1 and 269.66 gC m^-2.a^-1, respectively. Aspect reaches its highest NPP value in the fifth classification interval, with a value of 224.63 gC m^-2.a^-1.

Elevation.

Vegetation NPP varies with elevation, reflecting the ecological differences across the Tibetan Plateau. This study, based on 50 m elevation intervals, used GEE to illustrate NPP changes with altitude (Fig 9). As shown in Fig 9, NPP negatively correlates with elevation; as altitude increases, NPP decreases. This is linked to environmental factors like temperature, oxygen levels, and moisture availability [2].

Download:

• PPT
  PowerPoint slide
• PNG
  larger image
• TIFF
  original image

Fig 9. Relationship between vegetation NPP and elevation on the Tibetan Plateau from 2001 to 2021.

https://doi.org/10.1371/journal.pone.0320370.g009

At 500 m, the mean NPP is 145.017 gC m².a^-1, the highest in the study area (Fig 9). As elevation increases, NPP drops significantly to around 83.041 gC m².a^-1 at 2000 m, indicating a decline in productivity. Between 1500 and 3000 m, the decline is gradual (Fig 8). Above 3000 m, the decrease accelerates, particularly above 4000 m, where NPP sharply declines with every 500 m increase, falling from 30.169 gC m².a^-1 to about 0.34 gC m².a^-1, indicating near-zero productivity (Fig 9). At elevations near 7000 m, NPP is almost zero, showing the extreme limitations of high-altitude environments on vegetation growth [69,70].

Therefore, lower-altitude areas benefit from more favorable climatic and soil conditions, supporting vegetation growth, while higher altitudes pose increasing challenges to vegetation productivity.

Total solar radiation.

The Tibetan Plateau, sensitive to global climate change, is strongly influenced by total solar radiation, which drives photosynthesis and vegetation NPP [71]. Due to its high altitude, low temperatures, and unique monsoon climate, the impact of solar radiation on NPP shows distinct patterns [72,73]. The partial correlation coefficient between NPP and solar radiation ranges from ‐0.89 to 0.85, indicating significant spatial variation with both positive and negative correlations (Fig 10A).

Download:

• PPT
  PowerPoint slide
• PNG
  larger image
• TIFF
  original image

Fig 10. Partial correlation coefficients between vegetation NPP changes and climatic factors, including total solar radiation

(A), annual average temperature (B), accumulated temperature above 0 °C (C), and annual average precipitation (D), on the Tibetan Plateau from 2001 to 2021.

https://doi.org/10.1371/journal.pone.0320370.g010

As shown in Fig 10A, 34.16% of the area shows a positive correlation (partial correlation coefficient > 0). Among these, 7.90% have coefficients between 0.3 and 0.6, 0.36% exceed 0.6, and 25.89% are between 0 and 0.3. Solar radiation notably enhances NPP, particularly in regions with strong radiation and favorable temperature and moisture, like southwestern Qinghai, northern Sichuan, and eastern Tibet [70,72–74].

In contrast, 65.84% of the area shows a negative correlation (partial correlation coefficient < 0), where increased solar radiation suppresses NPP. Areas with coefficients less than ‐0.6, between ‐0.6 and ‐0.3, and between ‐0.3 and 0 represent 2.45%, 24.09%, and 39.30%, respectively. This negative correlation is likely due to high temperatures that reduce photosynthetic efficiency and increase water evaporation, hindering plant growth. These areas are mostly found in northeastern, eastern, and southwestern Qinghai, and parts of Tibet.

Annual average temperature.

Partial correlation analysis shows that the correlation between vegetation NPP and annual average temperature on the Tibetan Plateau ranges from ‐0.80 to 0.89, with both positive and negative correlations (Fig 10B). This aligns with studies highlighting temperature’s importance for plant growth but showing significant regional variation [75,76], contrasting with conclusions suggesting a general positive correlation without considering spatial heterogeneity [77].

Positive correlation: As shown in Fig 10B, 31.84% of the area shows a positive correlation (partial correlation coefficient > 0), with 30.02% having coefficients between 0.3 and 0.6 and 1.82% greater than 0.6. Temperature significantly promotes NPP in areas mainly in northeastern Qinghai, Tibet, and Sichuan. However, 52.54% of the area has a weaker effect (0 to 0.3), indicating temperature is not the primary driver of NPP variation in these regions, which are mainly in southern Qinghai, northwestern Sichuan, and Tibet.

Negative correlation: The negative correlation areas (Fig 10B) suggest that rising temperatures suppress NPP. Regions with coefficients below ‐0.5 account for 0.11%, and between ‐0.5 and ‐0.2, 2.98%, indicating high temperatures inhibit growth due to moisture loss. These areas are mainly in southwestern Tibet and northwestern Sichuan. Additionally, 12.53% of the area has a minor negative effect (between ‐0.2 and 0), suggesting other factors, such as precipitation or solar radiation, may play a larger role in NPP variation [10,78].

Accumulated temperature above 0 °C.

The partial correlation coefficient between vegetation NPP and accumulated temperature above 0 °C on the Tibetan Plateau ranges from ‐0.96 to 0.74, showing significant regional variation (Fig 10C).

Positive correlation: Positive correlation areas account for 15.22% of the total area. Of these, regions with coefficients between 0.2 and 0.5, and 0.5 to 0.74, make up 4.23% and 0.16%, respectively. These areas, mainly in the eastern and southeastern Tibetan Plateau and around the Sichuan Basin, show that increased accumulated temperature boosts plant photosynthesis, enhancing NPP under warmer conditions [11,12,79].

Negative correlation: Negative correlation areas cover 84.78% of the total area. Specifically, areas with coefficients below ‐0.7, between ‐0.7 and ‐0.4, and between ‐0.4 and 0 make up 6.91%, 27.21%, and 50.66%, respectively. In these regions, higher accumulated temperature increases evapotranspiration, reducing plant growth, particularly in the highlands of the Tibetan Plateau[7,80]. While moderate accumulated temperature promotes NPP, excessively high temperatures or water shortages hinder growth [81].

Annual precipitation.

The relationship between precipitation and NPP on the Tibetan Plateau is complex and region-dependent, with nonlinear characteristics [72]. The partial correlation coefficient between NPP and annual precipitation ranges from ‐0.89 to 0.90, showing both positive and negative correlations with significant spatial heterogeneity (Fig 10D).

Negative correlation: In negatively correlated areas, regions with coefficients below ‐0.6, between ‐0.6 and ‐0.3, and between ‐0.3 and 0 make up 1.26%, 15.20%, and 29.33%, respectively. Excessive precipitation may hinder plant growth, with other factors like temperature and radiation playing a more significant role [5,73]. These areas are mainly in eastern and southeastern Tibet, and northwestern Sichuan and Yunnan.

Positive correlation: In positively correlated areas, regions with coefficients between 0.3 and 0.6, and above 0.6 comprise 18.91% and 1.30%, respectively. Increased precipitation enhances NPP, especially in northeastern and southeastern Qinghai, northwestern Tibet, and southwestern Gansu [82].

Overall, while precipitation’s impact is limited in most areas, certain regions show stronger correlations, indicating that precipitation can both promote and inhibit vegetation growth.

Distance factors.

Distance from forest areas. Vegetation NPP decreases sharply as the distance from forest areas increases, then stabilises with minor fluctuations in areas farther away (Fig 11A). This suggests a significant positive influence of forest areas on nearby vegetation productivity, which weakens with distance [67]. As shown in Fig 10A, within 20,000 m of forest areas, NPP declines significantly, indicating higher productivity closer to forests due to benefits like improved soil moisture and nutrients, along with reduced human interference. Between 20,000 and 120,000 m, NPP stabilises at a lower level, showing a reduced forest influence. Beyond 120,000 m, fluctuations in NPP suggest influences from other factors, such as land use, human activities, or ecological variations.

Download:

• PPT
  PowerPoint slide
• PNG
  larger image
• TIFF
  original image

Fig 11. Variation in vegetation NPP with distance from forest areas (A) and roads(B).

https://doi.org/10.1371/journal.pone.0320370.g011

Distance from roads. The relationship between NPP and distance from roads follows a trend of initial decline, then a rise, and finally a sharp decrease, reflecting the impact of roads on vegetation (Fig 11B).Within 10,000 m of roads, NPP drops from 250 gC m^-2·a^-1 to 100 gC m^-2·a^-1, suggesting higher productivity near roads due to factors like accessibility and management. Between 10,000 and 50,000 m, NPP rises to 250 gC m^-2·a^-1, likely due to reduced disturbance and ecosystem recovery. Beyond 50,000 m, NPP decreases sharply to below 50 gC m^-2·a^-1, stabilising at 30 gC m^-2·a^-1 in areas over 70,000 m from roads, where harsh environmental conditions or minimal human activity limit vegetation growth [46].

Population density.

The correlation between NPP and population density on the Tibetan Plateau is ‐0.37, indicating a negative relationship. High population density areas near arable land and settlements often face vegetation degradation, soil erosion, and reduced biodiversity due to intensive human activities like grazing and agriculture, leading to lower NPP [83,84]. In contrast, low population density areas, such as nature reserves, show higher NPP due to minimal human disturbance and natural vegetation recovery [85].

Terrain and climate variations also influence this relationship. In high-altitude regions, harsh conditions result in low NPP despite sparse populations. Conversely, regions with moderate elevation and favourable conditions maintain higher NPP through effective management and restoration, even with higher population densities [86]. Some densely populated agricultural clusters achieve higher NPP via intensive practices and vegetation restoration [87].

Rising population density and infrastructure expansion increase resource demand, reducing vegetation cover and ecosystem connectivity, which further lowers NPP [28]. However, recent ecological policies and technological advancements have started reversing NPP declines in some high-density areas [88].

GDP density.

GDP density on the Tibetan Plateau, reflecting economic output per unit area, is closely linked to urbanisation, land use change, and resource exploitation, which directly and indirectly impact NPP [25,28]. Economic activities are concentrated in cities like Lhasa, Shigatse, and Nyingchi, where high GDP density often corresponds to significant human disturbance and reduced NPP [89].

In high GDP density regions, urban expansion, infrastructure development, and agricultural intensification alter land use, reducing vegetation cover and lowering NPP [25]. Urbanisation often diminishes green spaces, while agricultural expansion can initially raise NPP but may degrade soil and reduce NPP over time. Resource exploitation, such as mining, further damages ecosystems through vegetation destruction, soil erosion, and pollution [30].

Additionally, economic activities increase greenhouse gas emissions, intensifying climate change and indirectly affecting NPP by altering precipitation patterns and temperatures. These changes can exacerbate drought and cause earlier snowmelt, negatively impacting NPP [28].

Land use change.

Land use intensity. As shown in Fig 12, NPP on the Tibetan Plateau has a non-linear relationship with land use intensity. Moderate human disturbance promotes NPP growth, while excessive disturbance inhibits it [90,91]. This highlights the need for rational land use and vegetation protection to sustain ecosystems [92,93,94].

Download:

• PPT
  PowerPoint slide
• PNG
  larger image
• TIFF
  original image

Fig 12. Relationship between vegetation NPP and land use intensity from 2001 to 2021.

https://doi.org/10.1371/journal.pone.0320370.g012

At low land use intensity (<120), both the mean and standard deviation of NPP remain low and stable. For instance, at an intensity of 100, the mean NPP is 52.29 gC m^-2·a^-1, with a standard deviation of 36.59. Moderate land use intensity (130–160) increases both metrics, indicating enhanced vegetation growth but greater variability. For example, at an intensity of 158, the mean NPP is 425.26 gCm^-2·a^-1, with a standard deviation of 259.19.

In high-intensity areas (170–200), the mean and variability of NPP rise further. At an intensity of 174, the mean NPP is 381.75 gC m^-2·a^-1, with a standard deviation of 229.92. However, extremely high intensities (>200) show diverse trends: some areas see increased NPP with high variability (e.g., at 202, mean NPP is 579.02 gC m^-2·a^-1), while others experience declines (e.g., at 238, mean NPP drops to 160.19 gC m^-2·a^-1). This suggests that excessive activities, such as urbanisation and industrial expansion, negatively affect vegetation productivity [95].

Land use conversion. Land use conversion significantly influences carbon storage and ecosystem services on the Tibetan Plateau. Fig 13A shows major transitions from 2001 to 2021, with frequent exchanges between grassland and unused land. Grassland transformed into unused land, while unused land reverted to grassland and forest. Transitions between cropland and forest were less common, reflecting relative stability, and these changes are driven by human activities, environmental factors, and policies [67,96].

Download:

• PPT
  PowerPoint slide
• PNG
  larger image
• TIFF
  original image

Fig 13. Land use transitions (A) and corresponding vegetation NPP changes (B) on the Tibetan Plateau from 2001 to 2021.

Note: 2001 and 2021 in the figure represent the years 2001 and 2021, respectively.

https://doi.org/10.1371/journal.pone.0320370.g013

According to Fig 13B, converting 2,705 km² of cropland to forest increased NPP by 1.17×10⁶ tC, enhancing carbon storage and ecosystem services. Similarly, 4,680 km² of cropland converted to grassland raised NPP by 0.80×10⁶ tC, while 566 km² converted to water bodies altered NPP by 0.048×10⁶ tC, benefiting aquatic ecosystems [95]. However, 651 km² of cropland converted to built-up land decreased NPP by 0.095×10⁶ tC due to urban expansion [67,97].

Conversely, 2,425 km² of forest converted to cropland reduced NPP by 0.87×10⁶ tC, weakening carbon storage[98]. Similarly, 8,011 km² of grassland converted to cropland caused a significant NPP decline of 2.90×10⁶ tC, reflecting the adverse effects of agricultural expansion[99]. Other notable changes include 337 km² of water bodies converted to cropland, altering NPP by 0.12×10⁶ tC, and 362 km² of unused land converted to cropland, increasing NPP by 0.13×10⁶ tC, expanding agricultural areas [100,101].

Grazing. Grazing has a dual impact on vegetation NPP, with both benefits and drawbacks [92]. Moderate grazing reduces plant competition, enhances photosynthesis, improves soil fertility, and stabilizes grassland ecosystems [94]. In contrast, overgrazing causes vegetation degradation, soil damage, and NPP decline, replacing high-productivity plants with low-productivity species, and exacerbating soil compaction and erosion [91,93]. Climate change further interacts with grazing, affecting NPP by altering temperatures and precipitation. While warming may extend the growing season, excessive grazing can offset these benefits by suppressing plant growth [95].

Other factors.

While individual factors such as proximity, river and road density, slope, aspect, soil moisture, and moderate evapotranspiration have limited explanatory power alone, their combined effects significantly influence vegetation NPP changes. NPP variation on the Tibetan Plateau results from complex interactions between natural and human factors.

Proximity to forests, grasslands, and water bodies generally corresponds to higher NPP due to favorable moisture and vegetation conditions, whereas proximity to arable land and built-up areas often indicates lower NPP due to human disturbances [93]. Dense road and river networks fragment ecosystems, reducing NPP [46]. Slope and aspect impact vegetation by affecting soil erosion and solar radiation, with south-facing slopes usually showing higher NPP. Moderate soil moisture and evapotranspiration promote vegetation growth, but excessive evapotranspiration may inhibit NPP.

Similarities.

Consistent with previous studies, we focus on the Tibetan Plateau, analyzing the spatial and temporal variations in vegetation NPP. Similar to Xu et al. [69] and Piao et al. [2], we identify fluctuating upward trends in NPP and regional differences, with higher values in the southeast and lower values in the northwest cold desert, as noted by Zhang et al. [20]. Key drivers such as solar radiation, temperature, evapotranspiration, soil conditions, land use changes, and population density are confirmed in our research, aligning with studies like Sun et al. [102].

Differences.

Unlike prior studies, we applied advanced methods like Sen’s slope analysis, Hurst index, and Mann-Kendall test, combined with the OPGD model, to reveal non-linear relationships and spatial heterogeneity [103]. This enables us to quantify the interactions between natural factors and human activities, offering new perspectives often overlooked in other research [104,105].

Our regional subdivision analysis delves deeper into northeastern, southeastern, northwestern, and southwestern areas, uncovering spatial heterogeneity and providing new insights for targeted ecological protection strategies [106]. By integrating multi-source remote sensing and environmental data on the GEE platform, we achieved large-scale, long-term analyses with greater precision, enhancing the reliability and applicability of our findings [107,108].

Data limitations.

While we integrated multi-source remote sensing and environmental data, spatial and temporal resolution constraints may have missed subtle ecological changes [107,109]. Inaccuracies in key environmental variables could also affect the results [110]. Additionally, the long-term lag effects of ecosystem changes mean the trends observed may not fully capture future ecological evolutions [59].

Model assumptions.

The methods used, including Sen’s trend analysis and the OPGD model, are effective for identifying trends and drivers. However, assumptions and parameter choices could impact the results’ precision, especially when handling non-linear and complex interactions [103–105].

Regional heterogeneity.

Although regional subdivision analysis was conducted, small-scale heterogeneity and microclimatic effects may not have been fully captured, particularly in alpine desert and mountainous areas where terrain and micro-environments significantly affect NPP [15,21,67].

In conclusion, these limitations highlight the need for higher-resolution data, optimized models, and a focus on socio-economic factors in future research to better understand NPP changes on the Tibetan Plateau and improve ecosystem management [95,107,108].