Introduction

Vegetation is a key component of terrestrial ecosystems, influencing global climate, carbon storage, surface energy budgets, and the water cycle. Analyzing time series of vegetation parameters has been widely employed to explore phenology1, assess ecosystem health2, and investigate global climate change3. The Normalized Difference Vegetation Index (NDVI), which represents the ratio of near-infrared to red surface reflectance, is the most widely used metric for monitoring vegetation status and productivity4,5. Detecting trends in NDVI time series is essential for identifying and quantifying changes in ecosystem properties on scales ranging from local to global6. Vegetation is highly sensitive to environmental changes, making it a critical focus of studies on vegetation growth trends and the underlying drivers of these changes.

The Tibetan Plateau, the world’s largest plateau with an average elevation of approximately 4,000 m above sea level, provides an ideal environment for studying long-term vegetation trends. Its complex climate, low population density, and limited human activity create a unique natural laboratory7. The Tibetan Plateau is characterized by varied terrain and diverse surface properties8. Elevation gradients not only reflect topographical complexity but also play a critical role in amplifying the effects of climate change on vegetation dynamics. In mountain systems such as the Tibetan Plateau, elevation directly shapes the distribution and functional characteristics of vegetation by regulating the vertical differentiation of temperature, precipitation, and radiation. Previous research has primarily focused on general NDVI trends on the Tibetan Plateau9,10, but fewer studies have addressed how these trends are affected by elevation gradients and climatic factors in tandem. Linear regression is frequently used to quantify overall NDVI trends11, but this method only detects simple trends, missing more complex patterns such as abrupt changes12. Methods like piecewise linear regression and breakpoint detection algorithms have been developed to overcome these limitations. Various techniques, including the Landsat-based Detection of Trends in Disturbance and Recovery (LandTrendr) method13, Breaks for Additive Season and Trend (BFAST) method14, Vegetation Change Tracker (VCT) method15, and Detecting Breakpoints and Estimating Segments in Trend (DBEST) method16, have been proposed for detecting changes in vegetation dynamics. However, despite the effectiveness of these methods, the interaction between climatic variables and elevation gradients on vegetation dynamics in large mountainous regions such as the Tibetan Plateau remains underexplored.

Climate change, especially changes in temperature and precipitation17, is a major driver of vegetation changes in ecologically vulnerable regions18. The impact of precipitation on vegetation varies across different temporal and spatial scales. Long-term changes in precipitation can lead to shifts in vegetation types and ecosystem composition, while seasonal variations affect phenology and coverage19. In general, higher precipitation promotes vegetation growth in water-limited ecosystems. However, increased vegetation can feed back into the climate system by enhancing evapotranspiration, which can lead to surface cooling and increased precipitation20. Temperature is another critical factor that influences vegetation by regulating photosynthesis and other growth processes21. While previous studies have identified significant correlations between temperature, precipitation, and vegetation changes on the Tibetan Plateau, their combined effects along elevation gradients are still not well understood.

In addition, studies have suggested that the interaction between temperature, precipitation, and vegetation is particularly sensitive to elevation in mountainous regions. Recent evidence shows that warming rates in mountainous regions may exceed those of plains, a phenomenon known as elevation-dependent warming, which is crucial for understanding abrupt vegetation changes22. Zhang et al. demonstrated that significant vegetation changes are concentrated in areas with slopes less than 20° and elevations between 3,000 and 5,000 m23. However, large-scale research combining changes in both temperature and precipitation with elevation gradients on vegetation activity is scarce, highlighting the need for further studies in this area.

In this context, using the Tibetan Plateau as a case study area, our primary objective was to analyze vegetation dynamics during the period from 1982 to 2015, based on reconstructed high-precision NDVI remote sensing data. Specifically, we aimed to: (1) create a new high-resolution NDVI dataset by integrating multiple NDVI data sources; (2) analyze trends and detect abrupt changes in vegetation dynamics on the Tibetan Plateau; and (3) examine how NDVI varies with elevation and responds to climatic factors. This study provides new insights by investigating the joint effects of climatic factors and elevation gradients on vegetation dynamics, which could enhance our understanding of ecosystem responses to climate change.

Materials and methods

Study area

The Tibetan Plateau, known for its mean elevation surpassing 4000 m above sea level, is situated in central Asia, spanning from 25°59′ N to 40°29′ N latitude and from 73°27′ E to 104°30′ E longitude (Fig. 1). Due to its vast geographical coverage and significant spatial heterogeneity in climate and topography, the Tibetan Plateau exhibits notable spatiotemporal variations in surface vegetation patterns. Its low human disturbance and distinct environmental conditions make it one of the last regions where elevational vegetation dynamics can be studied under predominantly natural conditions.

Fig. 1
figure 1

Topography of the Tibetan Plateau and spatial distribution of meteorological stations. The top insert represents the percentage of meteorological stations across different elevation gradients.

Full size image

Data sources

NDVI data

Currently, most NDVI research employ three commonly used NDVI products: AVHRR NDVI, SPOTVGT NDVI, and MODIS NDVI. However, MODIS NDVI has only been providing data since 2002 and therefore cannot support long time series studies of vegetation cover change. In addition, there are many missing values in MODIS data for the Tibetan Plateau region due to heavy cloud contamination. Therefore, we utilized AVHRR GIMMS NDVI3g (hereinafter referred to as NDVIGIMMS) covering the period from 1982 to 2015 and SPOT-VGT NDVI (hereafter referred to as NDVISPOT) spanning from 1999 to 2013 were utilized24. The NDVIGIMMS dataset, derived from Advanced Very High-Resolution Radiometer (AVHRR) data (https://ecocast.arc.nasa.gov/data/pub/gimms), is the longest and most comprehensive dataset used in this study. However, the NDVIGIMMS dataset exhibits a nonstationary NDVI sequence, with a temporal resolution of 15 days and a spatial resolution of 1/12° (approximately 8 km at the equator), which limits its application in monitoring vegetation dynamics. The NDVISPOT dataset, characterized by higher quality but a shorter time series, spans from 1998 to 2013, with a spatial resolution of 1 km and a temporal resolution of 10 days (http://www.spot-vegetation.com). To mitigate the effects of geometric distortions, clouds, aerosols, and other factors, the maximum value composite (MVC) method was employed to calculate monthly maximum NDVIGIMMS and NDVISPOT values25.

Climatic dataset

Climate data were used to identify the potential factors influencing greenness trends before and after the breakpoints. The meteorological datasets included mean annual precipitation and temperature. These datasets were derived from daily observations collected at 134 meteorological stations strategically distributed across the Tibetan Plateau. The data source is the Chinese Meteorological Elements Day-by-Day Station Observation Dataset,’ made available by the Resource and Environment Science and Data Center of the Chinese Academy of Sciences (https://www.resdc.cn). Annual averages were obtained by summing daily temperature values and dividing by the number of days in the year. Similarly, the annual total precipitation was calculated by summing daily precipitation values over the entire year. The spatial distribution of the 134 meteorological stations across the Tibetan Plateau is shown in Fig. 1. These stations are predominantly concentrated in the eastern and southern parts of the Tibetan Plateau, with sparse distribution in the western region.

Elevation dataset

The digital elevation model data set (DEM) utilized in this study was obtained from the China Geospatial Data Cloud (http://www.gscloud.cn/). The DEM data, initially at a 90 m spatial resolution, was resampled to a 1 km resolution to align with the study’s spatial requirements.

Empirical orthogonal teleconnections

We employed the Empirical Orthogonal Teleconnections (EOT) model to fuse and generate high spatial resolution NDVI data26. The EOT model, based on multiple linear regression, calculates an empirical intersection function from a spatiotemporal dataset, yielding an orthogonal solution in the spatial or temporal ___domain. The updated NDVI dataset (referred to as NDVIEOT) generated from this analysis was used in subsequent stages of the study. We focused on NDVIEOT during the plant growing season (May to September) from 1982 to 2015 to minimize the effects of snow and ice27. The monthly maximum NDVIEOT values (referred to as \(\:{\text{NDVI}}_{\text{max}}\)) were derived from the corresponding NDVIEOT data, representing the peak NDVI during the growing season from 1982 to 2015.

The EOT analysis involves a stepwise linear regression where both predictors and outcomes are denoted by \(\:T\left(s,t\right),1\le\:t\le\:nt,1\le\:s\le\:ns\), where T represents the variable of interest. The process entails scanning all locations (s) to identify the base point (s, b) that best explains the variance at all other locations, including itself. Through standard regression, the explanatory components of \(\:T\left(s,t\right)\) are subtracted from \(\:T\left(sb,t\right)\), allowing for the identification of the subsequent most significant spatial point in terms of explaining variance. The final result is expressed as \(\:T\left(s,t\right)=\sum\:{\alpha\:}_{m}\left(t\right){e}_{m}\left(s\right)\), where the e’s represent spatial patterns, the α’s are time series, and the summation is over \(\:m=1,\dots\:,ns\).

The first EOT is intricately linked to a specific point, designated as sb1, wherein the time series \(\:{[\alpha\:}_{1}\left(t\right)]\) of the first mode corresponds to the time series of the original data at point \(\:s{b}_{1}\), denoted as \(\:T\left(s{b}_{1},t\right)\). Additionally, the space pattern of the first mode is the regression coefficient between \(\:T\) at \(\:s{b}_{1}\) and \(\:T\) at all other points \(\:s\).The EOT analysis was conducted using the ‘remote’ package in R version 3.6.3.

Piecewise linear regression method

Piecewise Linear Regression (PLR) segments time series data into adjacent linear sections based on temporal patterns, ensuring that the fitting error remains below a predefined threshold or matches a specific number of segments12. This method was applied to the time series of NDVImax to calculate the greenness rate of change during the growing season and to identify abrupt changes. The PLR model fitting was carried out using the ‘segmented’ package in R.

Breakpoint detection algorithm

The BFAST algorithm is vital for detecting both abrupt and gradual changes in seasonal time series data6. BFAST iteratively decomposes time series data into seasonal, trend, and residual components. The general model form is given by:

$$\:{Y}_{t}={T}_{t}+{S}_{t}+{e}_{t}(t=\text{1,2},\dots\:,n)$$
(1)

where \(\:{Y}_{t}\) symbolizes the identifiable data at time \(\:t\); \(\:{T}_{t}\) represents the trend component; \(\:{S}_{t}\) expresses the seasonal component; \(\:{e}_{t}\) is the remaining component that accounts for fluctuations not explained by \(\:{T}_{t}\) and \(\:{S}_{t}\); and \(\:t\) is the observation time.

\(\:{T}_{t}\) was assumed to be piecewise linear, with segment-specific slopes and intercepts across m + 1 segments. Thus, there are m breakpoints (\(\:{\tau\:}_{1},\dots\:,{\tau\:}_{m}\)), resulting in:

$$\:{T}_{t}={\alpha\:}_{i}+{\beta\:}_{i}t$$
(2)

where \(\:{\tau\:}_{0}\) = 0 and \(\:{\tau\:}_{m+1}=n\) are defined, and \(\:i=1,\dots\:,m\), The intercept and slope of the linear model are denoted by \(\:{\alpha\:}_{i}\) and \(\:{\beta\:}_{i}\), respectively.

The seasonal component, \(\:{S}_{t}\), denotes a piecewise harmonic model. For the segment in \(\:{t}_{j}<t<{t}_{j+1}\), set \(\:{t}_{0}=0\) and fit the seasonal component as follows:

$$\:{S}_{t}=\sum\:_{k=1}^{j}{\alpha\:}_{j,k}\text{s}\text{i}\text{n}(\frac{2\pi\:kt}{f}+{\delta\:}_{j,k})$$
(3)

where \(\:j\) defines the position of the breakpoints \(\:(j=1,\dots\:,m)\); \(\:k\) represents the number of harmonic terms; \(\:{\alpha\:}_{j,k}\) and \(\:{\delta\:}_{j,k}\) express segment-specific amplitude and phase, respectively; and \(\:f\) is the frequency.

The entire study period was divided into two segments: before and after the first detected breakpoint. The BFAST algorithm, implemented using the R package greenbrown, was used to identify the breakpoints and calculate trend slopes for the two periods.

Pearson correlation analysis

In this paper, we used the Pearson correlation coefficient to assess the influence of precipitation and temperature on \(\:{\text{NDVI}}_{\text{max}}\) from 1982 to 2015. The Pearson correlation coefficient (\(\:{r}_{xy}\)) is calculated using the following formula:

$$\:{r}_{xy}=\frac{\sum\:_{i=1}^{n}({x}_{i}-\stackrel{-}{x})({y}_{i}-\stackrel{-}{y})}{\sqrt{\sum\:_{i=1}^{n}{({x}_{i}-\stackrel{-}{x})}^{2}\sqrt{\sum\:_{i=1}^{n}{({y}_{i}-\stackrel{-}{y})}^{2}}}}$$
(4)

where \(\:n\) represents the number of data pairings, \(\:{x}_{i}\) and \(\:{y}_{i}\) are the values of the paired data, \(\:\stackrel{-}{x}\) and \(\:\stackrel{-}{y}\) denote the average values of each set of data, and \(\:{r}_{xy}\), ranging from − 1 to 1, indicates the degree of consistency between the two variables. As the absolute value of \(\:{r}_{xy}\) approaches one, it signifies a higher level of consistency between the two variables.

Results

Major shifts in vegetation greenness trends

To assess the accuracy of the synthetic NDVIEOT time series data, we employed a validation method encompassing both spatial and temporal dimensions. The performance of the EOT algorithm was evaluated by using 1998 as an independent validation year. As depicted in Fig. 2, the comparison between the mean NDVISPOT and the modelled NDVIEOT in 1998 shows that the EOT model achieved an R² value of 0.96 and a root mean square error (RMSE) of 0.001, indicating high accuracy in NDVIEOT prediction (Fig. 2d).

Fig. 2
figure 2

Maps showing (a) NDVI from SPOT, (b) NDVI from EOT in 1998, and (c) the resulting NDVI anomalies (△NDVI between NDVISPOT and NDVIEOT). White areas represent permanent snow and ice, while blue indicates water bodies. (d) Comparison of NDVI from SPOT and EOT in 1998 based on raster cells.

Full size image

Figure 3a illustrates the spatial distribution of the greenness rate of change of NDVImax during the growing season, using time-series data at a 1 km resolution. From 1982 to 2015, NDVImax displayed significant spatial variability, ranging from 0.09 to 0.89, with an average of 0.33. Areas with NDVImax values below 0.2, representing sparse vegetation typically found at high elevations, comprised approximately 21% of the study area. These regions are primarily located in the northwestern parts of the plateau, near the Altun, Himalaya, and Karakoram mountain ranges. The lower NDVImax values in these regions are attributed to sparse vegetation and high altitudes.

Conversely, regions where NDVImax ranged from 0.2 to 0.5 covered around 47% of the plateau, predominantly in the central areas near the Qaidam Basin and Tanggula Mountains. Vegetation in these areas includes alpine vegetation, grasslands, and deserts. Approximately 32% of the plateau exhibited NDVImax values exceeding 0.5, with these higher values concentrated in the eastern part of the Tibetan Plateau. Predominant vegetation types in these regions include primary forests, shrubs, and meadows, notably in the Sichuan-Tibet alpine canyon area, the Qinghai Plateau, the Qilian Mountains, and the Qaidam Basin.

Fig. 3
figure 3

Spatial pattern of the greenness rate of change on the Tibetan Plateau during the growing season from 1982 to 2015, based on NDVIEOT datasets, with significant changes (P < 0.05). Blue areas indicate water bodies.

Full size image

The spatial distribution of NDVImax trends from 1982 to 2015 exhibited considerable heterogeneity across the Tibetan Plateau (Fig. 3b). Over this period, NDVImax showed an average greening rate of 0.0012 per decade, highlighting a general increase in vegetation greenness across the Tibetan Plateau. The PLR method confirmed that the trend in growing season NDVImax was statistically significant throughout the study period. Approximately 86.47% of the area experienced a positive NDVImax slope (greenness rate of change > 0, p < 0.0001) between 1982 and 2015. This increase encompassed both modest gains (30.61%, 0.01 < greenness rate of change < 0.02) and substantial greening (26.23%, greenness rate of change > 0.02), whereas only 14% of the area showed browning trends (p < 0.0001). Over the last three decades, the Tibetan Plateau ‘s vegetation exhibited minimal change, with slight improvements and degradations (− 0.02 < greenness rate of change < 0.02) accounting for more than 93% of the total variation. In terms of different elevation gradients, the greening rate varies across elevation gradients, but overall shows an increasing trend. The NDVI increase is most pronounced below 1000 meters, at 1.42 × 10⁻³/year. The increase is slightly lower in the 1000–2000 meters and 3000–4000 meters ranges, at 1.32 × 10⁻³/year and 1.31 × 10⁻³/year, respectively. The increase in the 2000–3000 meters range is 0.88 × 10⁻³/year, which is relatively smaller compared to other elevation gradients.

A further analysis was conducted on NDVImax data obtained from 134 meteorological stations, situated at elevations ranging from 1,000 to over 4,000 m, and classified into four categories (Table 1). The average NDVImax across these stations was 0.54. About 8% of the stations recorded NDVImax below 0.2, while around 25% fell within the range of 0.2 to 0.5. Stations with NDVImax greater than 0.5 made up approximately 67% of the total. The relationship between NDVImax and elevation was as follows: stations located between 1,000 and 2,000 m had an average NDVImax of 0.66, those between 2,000 and 3,000 m averaged 0.56, stations between 3,000 and 4,000 m averaged 0.57, and those above 4,000 m averaged 0.41. NDVImax increased with elevation up to 4,000 m but declined at higher altitudes.

Table 1 Percentage of meteorological stations with mean NDVI values at different elevation ranges.
Full size table

Table 2 indicates that 96 stations exhibited an NDVImax change trend greater than 0, representing 71% of the total stations. NDVImax change trends by elevation are as follows: 0.0012 per decade for 1,000–2,000 m, 0.0017 for 2,000–3,000 m, 0.0018 for 3,000–4,000 m, and 0.0019 for elevations above 4,000 m. The data reveals a general increase in NDVImax change with higher elevation, where grasslands and meadows are more prevalent. Significant NDVImax change trends (greenness rate of change > 0.02) are notably more frequent at higher elevations, highlighting a pronounced greening effect in these regions.

Table 2 Percentage of meteorological stations with NDVI slopes at different elevation ranges.
Full size table

Using annual NDVImax data from 134 meteorological stations, we applied segmented linear regression to analyze the annual variation of NDVImax across different elevation gradients on the Tibetan Plateau from 1982 to 2015 (Fig. 4). The segmented regression indicated that NDVImax trends at these stations are heterogeneous, exhibiting non-stationarity and abrupt shifts. Specifically, breakpoints in the NDVImax trends occur earlier with increasing altitude. Our analysis identified a significant inflection point in 1998 (Fig. 4a). Prior to 1998, vegetation greenness increased modestly (slope = 0.0000038, R² = 0.07). After 1998, the greening trend became more pronounced (slope = 0.000032, R² = 0.74), signifying a notable shift in vegetation dynamics. Variation in NDVImax trends varied with elevation: altitudes above 2,000 m generally experienced increases in NDVImax, while areas between 1,000 and 2,000 m showed a decreasing trend. The 3,000–4,000 m elevation range demonstrated the most significant upward trend, coupled with relatively large fluctuations in NDVImax.

Fig. 4
figure 4

Long-term NDVI comparisons for different elevations: (a) Tibetan Plateau, (b) 1000–2000 m, (c) 2000–3000 m, (d) 3000–4000 m, (e) above 4000 m, based on 134 meteorological stations.

Full size image

Breakpoints in vegetation dynamics

Figure 5 illustrates the spatial patterns of breakpoints in NDVImax trends across the Tibetan Plateau, pinpointing the timing of significant changes. Abrupt changes are predominantly concentrated in the eastern regions, especially between elevations of 3,000 and 5,000 m. These areas show notable shifts in NDVImax trends, likely influenced by local climatic and environmental factors. The timing of these changes varies, with significant alterations observed in different years.

Fig. 5
figure 5

Ensemble of breakpoint and trend estimates from annual aggregated time series on the Tibetan Plateau. (a) Year of breakpoint detected in the NDVI time series. (b) Slope of the trend in the first segment. (c) Slope of the trend in the second segment. Circles represent meteorological stations.

Full size image

Figure 5a depicts the initial year of detected NDVImax breakpoints. From 1982 to 2015, approximately 19.31% of the Tibetan Plateau experienced NDVImax breakpoints. In these regions, the average greenness rate of change was 0.0014 per decade, with peak greenness rate of change reaching 0.0046 per decade in 1996 and 0.0036 per decade in 1993. Vegetation in 94–99% of the study area remained stable from 1982 to 1995. However, between 1996 and 2000, about 70.1% of vegetation areas experienced abrupt changes, predominantly in the eastern and southern regions, suggesting impacts from sudden climate shifts and warming. Notably, 59.4% of these regions saw their first breakpoint in 1998. Analysis by altitude indicates that the 4,000 to 5,000 m range contributes most significantly to the total mutation area, accounting for 44%, followed by the 3,000 to 4,000 m range at 24%. The mutation rate, defined as the proportion of mutation area to total vegetation area, is highest in the 2,000 to 3,000 m range at 27%, with the 3,000 to 4,000 m range also notably high at 25%. Lower mutation rates are observed at elevations of 1,000 to 2,000 m and above 4,000 m, both at 17%.

Figure 5b and c depict NDVImax trend changes before and after breakpoints, respectively. Prior to the breakpoints, most regions exhibited a positive greening trend with increasing NDVImax values. After the breakpoints, a majority of regions showed a negative browning trend with decreasing NDVImax values. Before the breakpoints, 77.95% of the study area experienced an upward trend, indicating increased vegetation greenness. Conversely, 22.05% of the area experienced a downward trend. Positive trends were mainly in the eastern Tibetan Plateau, while negative trends were observed in the central and southwestern regions. Post-breakpoint, only 15.48% of areas continued to show positive trends, with 84.52% exhibiting negative trends, suggesting widespread vegetation decline. Positive trends were largely confined to the Yellow River basin, indicating localized continued vegetation growth amid broader declines.

Further analysis of data from 134 meteorological stations revealed 41 sites with breakpoints, mainly located near rivers and lakes. About 49% of these stations had their first breakpoint in 1998. The average trend slope before the breakpoint was 0.00078 per decade, decreasing to 0.00049 per decade afterward, indicating a reduced rate of vegetation greening post-breakpoint. Before the breakpoints, 81% of the stations showed a positive NDVImax trend, and this proportion remained stable at 80% after the breakpoints, suggesting continued greening at most locations despite the detected shifts. Additionally, 71% of the stations exhibited consistent trend patterns across breakpoints, with trend slopes typically ranging from − 0.001 to 0.002 per decade, indicating stability in vegetation change rates over time.

Table 3 presents the distribution of abrupt changes detected at 41 meteorological stations across different elevation ranges between 1982 and 2015. The analysis reveals notable differences in breakpoint concentrations across elevation gradients. The year 1998 saw the highest number of breakpoints, with 36.6% of these occurring at elevations between 3,000 and 4,000 m above sea level. This is followed by 19.5% of breakpoints at elevations above 4,000 m and 14.6% at elevations between 2,000 and 3,000 m.

Table 3 Percentage of meteorological stations with abrupt changes in NDVI at different elevation ranges.
Full size table

Dynamics of Climatic factors

Figure 6 illustrates the interannual trends in precipitation and temperature across the 134 meteorological stations on the Tibetan Plateau from 1982 to 2015. The annual average temperatures on the Tibetan Plateau showed a consistent upward trend, ranging from 2.9 °C to 5.3 °C over the study period. Conversely, annual precipitation exhibited a fluctuating pattern, initially increasing and then decreasing, with values varying between 392 mm and 521 mm. Precipitation peaked at 521 mm in 1998, the highest value recorded during the 34-year study period. This year also saw significant increases in both precipitation and temperature, with rises of 24% and 59%, respectively, compared to the previous year. Prior to 1998, anomalous precipitation displayed a decreasing trend, with the rate of decline accelerating after 1998, although fluctuations remained high. Temperature trends were more pronounced, with an overall increase across all altitude gradients. Precipitation patterns, however, varied by altitude: a decreasing trend was observed at 1,000–2,000 m, a non-significant increase at 2,000–3,000 m, and a pattern of initial increase followed by decrease above 3,000 m.

Segmented trend analysis indicated that the 1,000–2,000 m range was unique in not experiencing abrupt changes in precipitation and temperature between 1997 and 1998. In contrast, other altitude ranges saw significant increases in both precipitation and temperature in 1998. Notably, the most substantial increase in precipitation occurred at the 3,000–4,000 m range, where it rose by 128 mm. Similarly, temperatures above 4,000 m increased by 1.8 °C, marking the highest rise among all elevation gradients.

Fig. 6
figure 6

Climate change trends in the time series data from 134 meteorological stations on the Tibetan Plateau.

Full size image

The response of NDVI to Climatic factor

To evaluate the impact of climatic factors on vegetation dynamics, we performed a correlation analysis between annual average NDVImax during the growing season and key climatic factors, as shown in Fig. 7. The analysis reveals a positive correlation between both precipitation and temperature with NDVImax across the Tibetan Plateau, highlighting their significant influence on vegetation. Specifically, NDVImax exhibits a clear linear relationship with precipitation, with 70% of meteorological stations showing a positive correlation (Fig. 7a). This relationship varies by elevation, but overall, NDVImax tends to increase with higher precipitation, supporting the positive correlation observed.

Similarly, a linear relationship is observed between NDVImax and temperature, with 72% of meteorological stations demonstrating a positive correlation (Fig. 7b). However, this correlation is somewhat weaker compared to that with precipitation. At elevations between 1,000 and 3,000 m, there is a positive correlation between NDVImax and temperature. In contrast, at elevations above 3,000 m, the relationship between NDVImax and temperature becomes less clear, with no significant linear or nonlinear pattern detected, suggesting a more complex interaction at higher elevations (Fig. 7i and j).

Fig. 7
figure 7

Scatter plots showing the relationship between NDVI, precipitation, and temperature at various elevations based on 134 meteorological stations. (a) and (b) for the Tibetan Plateau; (c) and (g) for elevations 1000 to 2000 m; (d) and (h) for elevations 2000 to 3000 m; (e) and (i) for elevations 3000 to 4000 m; (f) and (j) for elevations above 4000 m.

Full size image

We analyzed the relationship between annual average NDVImax during the growing season and climatic factors (Fig. 8) to assess their impact on vegetation dynamics. NDVImax demonstrates a strong correlation with both precipitation and temperature across the 134 meteorological stations, indicating the significant role of these climatic factors in shaping vegetation patterns.

Notably, the correlation between temperature and NDVImax is stronger than that between precipitation and NDVImax on an annual scale. The mean correlation coefficient between NDVImax and precipitation is 0.22, while with temperature, it is higher at 0.29. Regional analysis across different elevation zones reveals that the relationship between NDVImax and precipitation is strongest above 4,000 m, with an average coefficient of 0.24. This is followed by 0.22 at 2,000–3,000 m, 0.18 at 3,000–4,000 m, and the lowest at 0.15 below 1,000 m. In contrast, the correlation between NDVImax and temperature consistently exceeds that with precipitation. The highest correlation with temperature is observed at 1,000–2,000 m (0.36), followed by 3,000–4,000 m (0.34), 2,000–3,000 m (0.30), and the lowest correlation above 4,000 m (0.29).

Fig. 8
figure 8

Spatial pattern of correlation coefficients among the NDVI and (a) precipitation, (b) temperature on the Tibetan Plateau at different elevation gradients.

Full size image

Discussion

Temporal and spatial variation characteristics of vegetation

Monitoring long-term vegetation changes and understanding their contributing factors are crucial for global change research, as these changes reflect the broader impacts of climate and human activities on ecosystems28. This study examines the annual spatial distribution and spatiotemporal variations of NDVI across the Tibetan Plateau over the past 34 years, leveraging fused data from the GIMMS and SPOT NDVI datasets to enhance temporal and spatial resolution.

Using segmented linear regression and breakpoint detection, we identified a consistent increasing trend in NDVI across the Tibetan Plateau from 1982 to 2015. This finding aligns with previous studies29, confirming the effectiveness of these methods for detecting long-term vegetation changes. The spatial distribution pattern reveals that NDVI increases from northwest to southeast, reflecting the Tibetan Plateau’s climatic gradient. The arid western regions, characterized by lower precipitation and harsher climatic conditions, are less conducive to vegetation growth compared to the more humid eastern areas30. While vegetation in the eastern and northern Tibetan Plateau shows improvement trends, the southern regions exhibit signs of degradation. This degradation is likely driven by higher population density and intensified human activities, such as overgrazing, agricultural expansion, and urbanization, which exacerbate land degradation and reduce vegetation cover.

Significant breakpoints in vegetation cover changes were identified, with approximately 70.1% of vegetated areas experiencing one or more breakpoints between 1996 and 2000. Notably, 59.4% of these areas experienced their first breakpoint in 1998, which may correspond to significant climatic shifts and the implementation of large-scale ecological restoration projects. This timing aligns with notable climatic and anthropogenic changes on the Tibetan Plateau29, including the introduction of major ecological restoration initiatives.

The observed trend is likely influenced by the large-scale ecological restoration projects initiated by the Chinese government after 1998, such as the Natural Forest Protection Program and the Grain for Green Project. These initiatives aimed to promote reforestation, reduce deforestation, and enhance soil conservation, significantly contributing to surface greening, particularly in the northwest region of China. The Natural Forest Protection Project began in 1998, and the policy of returning farmland to forest commenced in 1999. Subsequent ecological measures, including various reforestation and water conservancy projects, have contributed to the observed increase in regional NDVI. This may explain the large fluctuations in NDVI around 1998 and the more stable trends in other years31,32. The greening efforts in the northwest region of China validate our findings. These initiatives have not only enhanced vegetation recovery and regrowth but have also significantly reduced surface runoff and soil erosion, improving water use efficiency and ecosystem stability33,34,35.

Climatic factors behind changing patterns of vegetation dynamics

We integrated NDVI data with temperature, precipitation, and topography to assess their effects on vegetation across the Tibetan Plateau. Analysis of data from 134 meteorological stations reveals a significant increase in temperature and relatively stable precipitation levels. These trends have a crucial role in shaping vegetation dynamics. Correlation analysis shows a positive relationship between NDVI and both temperature and precipitation across most regions of the Tibetan Plateau, highlighting the role of these climatic factors in driving vegetation changes.

These findings are consistent with previous studies, though our analysis suggests a stronger influence of temperature than traditionally recognized. While precipitation has often been viewed as the dominant factor, our study indicates that temperature exerts a more critical influence on vegetation dynamics on the Tibetan Plateau, particularly in higher elevation regions. This finding is in line with Li et al.30, emphasizing the increasing relevance of temperature in driving vegetation responses under changing climatic conditions.

The stations display consistent NDVI trends corresponding to regional precipitation patterns, suggesting a rapid NDVImax response to changes in precipitation. This responsiveness highlights the sensitivity of vegetation to short-term climatic fluctuations, particularly in regions where water availability is a limiting factor. Notably, the abrupt change in precipitation and temperature in 1997, peaking in 1998, aligns with the sudden \(\:{\text{NDVI}}_{\text{max}}\) changes. Numerous studies report a lag of 20–40 days between NDVI and meteorological parameters, while others suggest a longer lag of about 2–3 months32.

The observed lag effects between NDVImax and climatic factors, along with spatial variations, reveal the intricate relationship between meteorological conditions and vegetation dynamics. This study highlights the profound impact of climate change on regional vegetation patterns, consistent with prior research36. Ongoing monitoring and further detailed studies are essential to understand the complex interactions between climate and vegetation, particularly under future climate scenarios.

Variation in characteristics of NDVI under different elevation

Elevation on the Tibetan Plateau plays a crucial role in shaping spatial patterns and climate dynamics, significantly influencing vegetation processes37. Understanding the impact of elevation on these dynamics is essential for predicting vegetation responses to climate change. As elevation is increasingly recognized as a limiting factor, its influence on large-scale vegetation dynamics becomes more pronounced under shifting climate conditions. Terrain features such as slope and aspect further contribute to variations in vegetation greenness trends across different elevations38. Previous studies have noted faster warming at higher elevations, affecting vegetation dynamics in mountainous regions39.

This study evaluates how elevation impacts climatic factors and NDVI breakpoints across 134 meteorological stations on the Tibetan Plateau, using elevation-stratified correlation analysis to assess variations in vegetation response at different altitudinal ranges. Our results indicate a stronger correlation between NDVI and climate factors, particularly temperature and precipitation, at higher elevations, highlighting the increased sensitivity of these regions to climatic changes40. The average correlation index rises with elevation, demonstrating the heightened sensitivity of high-altitude areas to climate variability. Regions between 3000 and 4000 m, primarily covered with grasslands and cultivated vegetation, experience significant abrupt changes. As elevation increases, precipitation generally rises while temperature declines, easing constraints on vegetation growth. Optimal hydrothermal conditions around 3000–4000 m lead to peak NDVImax, reflecting vegetation adaptation to these favorable conditions41. Regarding the phenomenon of the lowest greening rate within the 2000–3000 m altitude range, we can analyze it from multiple perspectives. First, although the correlation between temperature and NDVI is the highest in this range, this does not necessarily mean that an increase in temperature will lead to an increase in greening rate. There may be other constraining factors, such as limitations of precipitation or changes in soil conditions. Research has found that in high-altitude areas, precipitation often decreases with increasing elevation, which may lead to insufficient water availability for vegetation growth, thereby restricting the greening rate. The range of 2000–3000 m may be in a temperature-sensitive zone (where warming leads to drought stress) or a critical belt of precipitation redistribution, thereby triggering breakpoints in NDVI.

The unique geographical environment of the Tibetan Plateau also plays a significant role in influencing local vegetation. Our study shows a decrease in vegetation cover in the 4000–5000 m altitude range, contrasting with some studies that report increases in the 4500–5000 m range42. These differences may be due to local environmental conditions and dominant vegetation types in these areas. Future research should focus on vegetation changes within the 4000–5000 m range on the Tibetan Plateau.

The spatial variation of NDVImax across elevations on the Tibetan Plateau reflects the complex interplay between water availability and temperature. This nuanced understanding of elevation-related NDVI variations offers valuable insights into vegetation dynamics on the Tibetan Plateau and underscores the need for tailored climate adaptation and ecological conservation strategies. Long-term monitoring and modeling are essential for predicting vegetation responses under varying climatic scenarios, particularly in high-altitude regions.

Limitations of this study

This study has several limitations that should be acknowledged. Firstly, the focus was primarily on key climatic factors such as precipitation, temperature, and elevation, while other significant factors, including wind speed, solar radiation, and relative humidity, were not considered. These additional variables can substantially influence vegetation growth and dynamics, especially in the complex environment of the Tibetan Plateau. Future research should aim to incorporate these factors to achieve a more comprehensive understanding of their effects on vegetation changes. Employing advanced modeling techniques to integrate a broader range of environmental parameters could provide deeper insights into the drivers of vegetation dynamics on the Tibetan Plateau.

Secondly, although the study generated a high-resolution map of vegetation dynamics, there are limitations related to the data and methodologies used. The spatial and temporal resolution of satellite imagery may not fully capture short-term vegetation responses or fine-scale ecological variations. Additionally, cloud cover has posed a significant challenge to the accuracy of satellite observations, particularly in the southeast Tibetan Plateau where cloud cover is more prevalent43. This issue may lead to gaps in data or inaccuracies in monitoring vegetation. To enhance data reliability, future research could explore cloud-penetrating sensors, such as Synthetic Aperture Radar, or advanced cloud-masking algorithms. While multisensor data fusion was used to address gaps in data density, this approach may introduce uncertainties, particularly concerning the alignment and integration of data from different sensors44. Future studies could focus on refining fusion techniques or validating results with ground-based observations to minimize these uncertainties.

Conclusion

This study utilized the EOT algorithm to enhance NDVI accuracy by addressing spatial and temporal inconsistencies, demonstrating the high applicability of the spatiotemporal fusion-simulated NDVI dataset for monitoring vegetation trends on the Tibetan Plateau. From 1982 to 2015, the Tibetan Plateau exhibited a consistent greening trend, with 86% of the vegetation-covered area showing increased growth, while only 14% experienced browning. The average NDVI change rate is 0.0012 per decade, with higher values observed in the southeast and lower values in the northwest, reflecting regional climatic and topographical influences.

Significant greening is noted above 4000 m, with a rate of 0.00199 per decade, suggesting favorable conditions and reduced human impact. The altitudinal variation in NDVI increase underscores the complex interplay between elevation, climate, and vegetation dynamics. Stable vegetation covered 94–99% of the area each year, with 70.1% of vegetation areas showing abrupt changes between 1996 and 2000, particularly in the eastern and southern regions. Notably, 59.4% of these areas had their first breakpoint in 1998, indicating a significant shift in vegetation dynamics.

Elevation significantly affects breakpoint timing, with the highest NDVI breakpoint rates between 2000 and 3000 m (27%) and lower rates at other elevations. Climatic factors, particularly temperature, strongly influence NDVImax anomalies, with temperature having a more pronounced effect than precipitation. The impact of climate change on vegetation trends is further amplified by terrain characteristics, with breakpoint sites concentrated above 4000 m. Overall, the influence of climatic factors on vegetation varies with altitude, highlighting the need for continued monitoring and targeted conservation efforts to sustain ecological improvements amid ongoing climate change.