Development of long‐term spatiotemporal continuous NDVI products for alpine grassland from 1982 to 2020 in the Qinghai–Tibet Plateau, China

Abstract

Background

The time‐series data of the Normalized Difference Vegetation Index (NDVI) is a crucial indicator for global and regional vegetation monitoring. However, the current assessment of global and regional long‐term vegetation changes is subject to large uncertainties due to the lack of spatiotemporally continuous time‐series data sets.

Methods

In this study, a long time‐series monthly NDVI data set with a spatial resolution of 250 m from 1982 to 2020 was developed by combining Moderate Resolution Imaging Spectroradiometer (MODIS) and AVHRR (Advanced Very High‐Resolution Radiometer) time‐series NDVI products using the Random Forest (RF) downscaling model.

Results

Compared to the MODIS NDVI product, the fused product shows RMSE and mean absolute error ranging from 0 to 0.075 and from 0 to 0.05, respectively, with R² values mostly above 0.7.

Conclusions

The long time‐series NDVI products generated in this study are reliable in terms of accuracy and have great potential for long‐term dynamic monitoring of terrestrial ecosystems on the Qinghai–Tibet Plateau.

INTRODUCTION

In recent years, global climate change, especially warming, has had a significant impact on ecosystems (Landman, 2010; Li et al., 2018). The Qinghai–Tibet Plateau, a warning and sensitive area for global warming (Liu & Chen, 2000), has experienced rapid climate warming over the past few decades, with a rate of warming twice the global average (You et al., 2016). The ongoing changes in global climate will exacerbate the impact on ecosystems (Li et al., 2018; Warren et al., 2011; Wu et al., 2019). Satellite-based remote sensing has become the most commonly used monitoring vegetation dynamics due to its wide coverage and strong spatial and temporal resolution characteristics (An et al., 2015). The Normalized Difference Vegetation Index (NDVI), a metric that quantifies vegetation's absorption and reflection properties (Huete et al., 2002; Tucker, 1979), has gained widespread adoption for the assessment of vegetation dynamics due to its straightforward computation, high sensitivity in vegetation surveillance, and robust spatiotemporal flexibility, rendering it an essential tool for evaluating regional and global vegetation alterations (Cui & Shi, 2010; Jiang et al., 2006; Lan & Dong, 2022; Martinez & Labib, 2023).

Currently, the primary long-term vegetation index data sets include the Global Inventory Monitoring and Modeling System (GIMMS) and Moderate Resolution Imaging Spectroradiometer (MODIS) NDVI products. The GIMMS NDVI data set (1981–2015), derived from the AVHRR (Advanced Very High-Resolution Radiometer) sensors aboard the NOAA satellite series, offers a 15-day temporal resolution and an 8-km spatial resolution. It effectively mitigates the influences of volcanic eruptions, solar zenith angle variations, and sensor sensitivity alterations over time, thus enjoying widespread global application (Pinzon & Tucker, 2014; Tucker et al., 2005). However, due to its lower spatial resolution, it cannot capture the fine-scale details required for heterogeneous ecosystems and land cover change (An et al., 2018; Chen et al., 2021). Conversely, the MODIS NDVI (2000–present) data are considered an improved version of the AVHRR NDVI data, with enhanced spatial resolution and chlorophyll sensitivity. It eliminates water vapor disturbances, reduces radiometric distortion, and adjusts synthesis methods (van Leeuwen et al., 2006). Furthermore, the Landsat satellite series is capable of offering long-term, high-resolution vegetation index observations dating back to 1972. However, its effectiveness is hindered by a 16-day revisit period and susceptibility to significant cloud cover (Anderson et al., 2021; Sano et al., 2007), leading to a reduced number of clear observations in a given year (Cao et al., 2020; Ju & Roy, 2008; Shen, Li, et al., 2015; Shen, Piao, et al., 2015).

Previous studies have indicated that the spatial patterns and dynamic tendencies of GIMMS NDVI and MODIS NDVI within the Qinghai–Tibet Plateau are largely consistent (Du et al., 2014). A high pixel correlation between these data sets suggests their efficacy in similarly capturing vegetation growth cycles (Fensholt & Proud, 2012). Within the Qinghai–Tibet Plateau, the correlation and discrepancies between these data sets are more discernible compared to other arid regions globally (Fensholt et al., 2012; Ma et al., 2022). Differences still exist among data sets in terms of spatial and temporal scales (Marshall et al., 2016; Zhou et al., 2015). The reliance on a single data source poses challenges for long-term sequence analyses, leading to ongoing debates in research focused on monitoring vegetation dynamics through vegetation index products (Ding et al., 2015; Nagol et al., 2014; Zhang et al., 2013). In earlier studies, numerous scholars utilized GIMMS NDVI products to explore vegetation change trends, revealing a stagnation or even reversal of the global greening trend after 2000 (Yuan et al., 2019). MODIS and SPOT data show significant greening from 2000 to 2015, while GIMMS shows noticeable browning in the Tibetan Plateau (Ding et al., 2007; Liu et al., 2022; Zhang, Cheng, et al., 2022; Zhong et al., 2019; Zhou et al., 2007). Additionally, a study has found browning in the southwestern part of the plateau during the growing season in the 2000s (Shen, Li, et al., 2015; Shen, Piao, et al., 2015). However, a correlation study utilizing MODIS products discovered a reversal from browning to greening in NDVI from the 2000s to the 2010s (Li et al., 2020). Currently, there are two approaches: one involves calibrating the NDVI based on data acquired when there is no orbital drift in the NOAA or when the AVHRR sensor is not degraded (Jiang et al., 2008; Pinzon & Tucker, 2014), and the other involves calibrating the AVHRR NDVI using other sensors with overlapping observing periods (Cao et al., 2004; Tucker et al., 2005). However, accurately calibrating NDVI products is challenging without fully understanding the product accuracy of each period and considering the dynamic changes of calibration model parameters. To address this, we are dedicated to creating a set of continuous, high-resolution NDVI products that improve the reliability and accuracy of current vegetation monitoring data, supporting a more precise understanding of trends in vegetation ecosystem change.

Data fusion algorithms contribute toward combining the strengths of different sensors, improving the spatiotemporal resolution of fused products in spatial, spectral, or temporal dimensions (Cheng et al., 2020; Lima et al., 2021; Moreno-Martinez et al., 2018; Zurita-Milla et al., 2009). The AVHRR NDVI product was downscaled using a geographically weighted linear mixture model at early times when the spatial resolution of the satellites was typically low. However, the differences in land cover types made the model produce significant uncertainty and resulted in substantial errors (Kerdiles & Grondona, 1995; Somers et al., 2011). The use of Spatiotemporal Adaptive Reflectance Fusion Model (STARFM) to fuse Landsat and MODIS data allows the production of time-series data with higher spatial resolution of surface reflectance (Gao et al., 2006). A long-term NDVI time series for the Three-River Headwaters Region can be obtained based on Landsat and MODIS data using an improved residual convolutional neural network model (Sun et al., 2024; Sun & Wang, 2022). Nevertheless, these downscaling methods may experience a decrease in estimation accuracy in deeply fragmented and heterogeneous patches (Cui et al., 2018). In addition, downscaled data may contain additional information than the original data, but possibly there is a loss of information and potential spatial registration errors in the conversion to fine-scale resolution (Yu et al., 2013) due to inaccuracies in data fusion, as well as nonnegligible errors (Ge et al., 2019; Qu et al., 2021).

With the continuous evolution of remote sensing image processing technologies, machine learning algorithms such as Random Forest (RF), Support Vector Machine (SVM), Gaussian Process Regression, Artificial Neural Networks, and so forth have shown excellent performance in scale transformation and spatiotemporal modeling of remote sensing data (Ali et al., 2021; Huang et al., 2023; Karbalaye Ghorbanpour et al., 2021; Liu et al., 2020; Sdraka et al., 2022; Yan et al., 2021). Among these, RF performs better in reconstructing NDVI time-series products than the Long Short-Term Memory Artificial Neural Networks (Sun, Gong, et al., 2023; Sun, Li, et al., 2023). Based on the RF model, the fusion of AMSR-E and MODIS data can effectively avoid excessive smoothing, providing land surface temperature (LST) estimates that better match real-world conditions (Zhang, He, et al., 2022); when downscaling evapotranspiration using three machine learning methods (RF, SVM, Cubist), the RF model produces the smallest error, demonstrating its potential in scale transformation (Ke et al., 2016). In addition, the RF model has better processing ability for high-dimensional data, and can contain more factors related to target variables to improve the explanatory and predictive accuracy of the model (Li et al., 2023).

Therefore, in this study, we use the RF algorithm for spatiotemporal modeling by analyzing the GIMMS and MODIS cross-period vegetation indices and taking into account auxiliary factors such as topography. The aim is to produce a set of monthly downscaled NDVI products for the Tibetan Plateau with a spatial resolution of 250 m covering the period from 1982 to 2020.

MATERIALS AND METHODS

Study area

The Qinghai–Tibet Plateau is located in southwestern China, with an average elevation of over 4000 m (Figure 1a). It is the world's largest and highest plateau, and is known as the “Roof of the World” (Ding et al., 2023). Within China, it stretches from the Pamir Plateau in the west to the Hengduan Mountains in the east, spanning approximately 31 degrees of longitude, with an area of approximately 2 572 400 km², accounting for 27% of China's land area (Yao et al., 2012). As an alpine climate zone, grassland vegetation predominates in this region, with main grassland types including alpine meadow, alpine steppe, and alpine desert steppe (Wu et al., 2007) (Figure 1b). It makes the plateau one of the most significant pastoral areas in China (Wu et al., 2007; Yuan et al., 2022). Therefore, generating a long-term, high-resolution NDVI data set is of great importance for monitoring the long-term vegetation changes in this region. It can provide a solid data basis for ecological and environmental monitoring and sustainable development in the region.

[IMAGE OMITTED. SEE PDF]

Data sources

In our study, the MODIS NDVI data (MOD13Q1) originate from the NASA MODIS Land Product suite, developed using a unified algorithm for MODIS Vegetation Index products. The data set, with a temporal resolution of 16 days and a spatial resolution of 250 m, was downloaded from the MODIS Web (). The product's time frame spans from February 28, 2000 to December 31, 2020. The MODIS Reprojection Tools were used to mosaic the data for each tile, with the projection defined as WGS 1984/Albers.

The GIMMS NDVI 3g product is from the AVHRR instrument positioned on the NOAA polar-orbiting meteorological satellite (). This product spans the period from 1982 to 2015 with a temporal resolution of 15 days and a spatial resolution of 8 km. It has been widely used globally, eliminating the effects of factors such as volcanic eruptions, solar zenith angle, and sensor sensitivity changes over time (Pinzon & Tucker, 2014; Tucker et al., 2005). While the two data sets show relatively consistent monthly and annual variability in vegetation, the timing of the NDVI peaks is significantly inconsistent (Figure 2), introducing uncertainty into the study of vegetation phenology. Therefore, to resolve the uncertainties caused by the differences between the data, we produced a long time-series, high-spatial-resolution data set based on the two aforementioned data sets.

[IMAGE OMITTED. SEE PDF]

The DEM data used in this study are provided by the Shuttle Radar Topography Mission (SRTM) operated by the National Geospatial-Intelligence Agency and the National Aeronautics and Space Administration (Rodriguez et al., 2006; Schneider & Reinartz, 2009). The available options for spatial resolution include both 30 m and 90 m, with the latter being accessible at . The DEM, longitude, and latitude can illustrate the hydrothermal conditions of a region on the Qinghai–Tibet Plateau, offering insights into zonal variations in vegetation and facilitating more precise predictions of spatial variation and vegetation distribution (Li et al., 2023).

Data preprocessing

The integration of disparate spatial data was accomplished through the application of the Albers Equal Area Conic Projection System and the WGS 1984 Coordinate System. The Maximum Value Composite method was applied to synthesize MODIS and GIMMS NDVI results into monthly time series for each pixel, which extracts the maximum NDVI value at the pixel level from all available NDVI observations within each month (Faisal et al., 2019). Subsequently, the nearest-neighbor method was used to resample the GIMMS data to 250 m, aligning its resolution with the MODIS NDVI. Similarly, we also used the nearest-neighbor method to resample the 90 m resolution DEM to 250 m. Finally, the interleaving period between 2000 and 2015 for the GIMMS and MODIS data sets was chosen for model construction and evaluation. In this process, monthly downscaled models are built using odd-year data and accuracy is evaluated using even-year data.

Random Forest

RF is an integrated learning method that improves the accuracy and stability of predictions by constructing multiple decision trees and voting or averaging them. Considering the unavailability of Landsat data on a daily basis, as well as their significant disparity in spatial resolution compared to GIMMS NDVI, we deem GIMMS NDVI and MODIS NDVI as ideal data sources for model construction due to their matching temporal resolution and overlapping time periods. During the model training phase, we screen auxiliary data that have a substantial impact on NDVI as regression variables to augment both the prediction accuracy and the explanatory power of our model. The NDVI downscaled model uses the RF machine learning techniques to establish a mapping relationship between a set of regression explanatory variables and the NDVI, resulting in high-spatial-resolution NDVI data. The specific steps involved are as follows: (1) constructing the RF downscaling model using MODIS NDVI as the target variable and GIMMS NDVI, DEM, longitude, and latitude as regression explanatory variables with the same resolution as training samples; (2) inputting different parameters and comparing the results of the test set and the training set to evaluate the model's performance and data fitting effect; and (3) adjusting model parameters to create different models, comparing their performance, and selecting the optimal model.

Accuracy assessment

Three standard indices including root mean square error (RMSE), mean absolute error (MAE), and coefficient of determination (R²), to validate the error between MODIS NDVI data and downscaled NDVI data, and a time frame of 84 months from even years in the period 2001–2015 was used to avoid data overfitting. The formula for each index is as follows: 1$RMSE=\sqrt{{\sum }_{i=1}^N{({\mathit{NDVI}}_p\left(i\right)-{\mathit{NDVI}}_s\left(i\right))}^2/N},$ 2$\mathrm{MAE}=\frac{1}{N}\sum _{i=1}^N\left|({\mathit{NDVI}}_p\left(i\right)-{\mathit{NDVI}}_t\left(i\right))\right|,$ 3$R^2=1-\frac{{\sum }_{i=1}^N{({\mathit{NDVI}}_p(i)-{\mathit{NDVI}}_s(i))}^2}{{\sum }_{i=1}^N{({\mathit{NDVI}}_p(i)-{\mathit{NDVI}}_t(i))}^2},$where NDVI_p(i) is the value of the MODIS NDVI product, NDVI_s(i) is the fitted value of the downscaled NDVI product, and N represents the number of pixels involved in the test.

Trend test

To validate the performance of the downscaled products in long-term vegetation monitoring, we use the Mann–Kendall test and Sen's slope estimation to compare the trends at the pixel level in the GIMMS NDVI, MODIS NDVI, and downscaled NDVI products. Sen's slope is a nonparametric statistical method for assessing time-series trends, which reduces the effect of outliers while accommodating significant data distributions. We compute the standardized test statistic Z-value to measure the trend in the time series and whether it is significant or not. At a given significance level α, if |Z| > Z_1−α/2, it indicates a significant shift in the time-series data at the α level. Typically, α is set to 0.05. Referring to the table, when |Z| > 1.96, the time-series confidence level α < 0.05, passing the significance test.

RESULTS

Model optimization

The model complexity is contained by the number of decision trees and maximum depth to avoid overfitting and underfitting. In our study, an optimal model comprising 10 decision trees and 10 leaf nodes was established for each month. Overall, all the available explanatory variables contribute to each downscaled model in this study, with RMSE ranging from 0.0644 to 0.0661, MAE ranging from 0.0389 to 0.0396, and R² exceeding 0.8980 in all cases (Table 1). It is noteworthy that due to the lush vegetation in the summer, the NDVI values are significantly higher (Wang et al., 2022), resulting in the downscaling models demonstrating lesser generalization capabilities in June, July, and August compared to other months. The monthly downscaled model shows neither overfitting nor underfitting, which demonstrates the feasibility of building RF-based downscaled models.

Table 1 Evaluation metrics for the downscaling model in each month.

Month	Train_RMSE	Test_RMSE	Train_MAE	Test_MAE	Train_R2	Test_R2
January	0.0539	0.0556	0.0318	0.0324	0.9027	0.8967
February	0.0517	0.0533	0.0302	0.0308	0.9043	0.8980
March	0.0509	0.0526	0.0292	0.0298	0.9032	0.8965
April	0.0544	0.0563	0.0306	0.0312	0.8993	0.8921
May	0.0661	0.0679	0.0392	0.0399	0.9006	0.8951
June	0.0805	0.0823	0.0507	0.0515	0.8974	0.8927
July	0.0853	0.0869	0.0543	0.0551	0.9032	0.8993
August	0.0847	0.0865	0.0538	0.0547	0.8973	0.8927
September	0.0665	0.0677	0.0424	0.0429	0.9238	0.9212
October	0.0654	0.0676	0.0386	0.0394	0.8942	0.8871
November	0.0575	0.0590	0.0341	0.0347	0.9106	0.9061
December	0.0555	0.0571	0.0322	0.0327	0.9054	0.9000
Overall	0.0644	0.0661	0.0389	0.0396	0.9035	0.8981

Accuracy assessment

Overall assessment

The validation results are shown in Figure 3. In 62.4% of the Qinghai–Tibet Plateau region, the RMSE is below 0.025 and is mainly found in the central and western parts of the plateau, where the vegetation consists mainly of alpine meadow and alpine steppe. Regions with RMSE between 0.025 and 0.1 account for 31.2% and are sparsely scattered in the eastern plateau, mountain valleys, and areas around Qinghai Lake, as well as some low-altitude areas with diverse grassland types, including alpine meadow, alpine steppe, and temperate steppe. Only 6.4% of the area has an RMSE above 0.1, and the predominant grassland type is mountain meadow. Regions with MAE above 0.1 are only 2.9% compared to the RMSE, mainly located in the southeastern part of the plateau, where mountain meadow is the main vegetation type. In addition, 43.4% of the plateau region has an R² higher than 0.7. The validation results indicate that the downscaling algorithm performs well in most grassland areas of the Qinghai–Tibet Plateau. Therefore, the long-term NDVI products produced in this study should be reliable for the evaluation of grassland dynamics in this area.

[IMAGE OMITTED. SEE PDF]

Monthly accuracy assessment

The scatter plots of the fitting density for each month show that the downscaled NDVI products demonstrate excellent accuracy in all metrics (R² > 0.809; RMSE < 0.113; MAE < 0.068) and most pixels are closely clustered around y = x (Figure 4). From January to April, NDVI concentrated within the range of 0 to 0.4. As the NDVI values increased, there was a trend of underestimation in downscaled NDVI, but the data accuracy still remained elevated (R²: 0.856–0.868; RMSE: 0.06–0.064; MAE: 0.034–0.037). During the vegetation growing season (May–September), NDVI varied in a wide range, within 0–1, but the color band showed that NDVI was mostly in the range of 0–0.2, with a trend from overestimated to underestimated. Moreover, compared to the previous 4 months, lower accuracy (R²: 0.809–0.875) and higher errors (RMSE: 0.077–0.11; MAE: 0.044–0.068) were found. Additionally, due to the presence of evergreen needleleaf forests in part of the Himalayas and the Hengduan Mountains of the Qinghai–Tibet Plateau (Cong et al., 2013; Ding et al., 2007; Lv et al., 2023), pixels with NDVI values exceeding 0.6 persisted during periods of reduced vegetation growth (e.g., January, February, March, April, October, November, and December).

[IMAGE OMITTED. SEE PDF]

Accuracy assessment in various grassland types

The downscaled product typically showed consistent NDVI trends with MODIS across nine grassland types, which are the main types that dominate the area of Qinghai–Tibet Plateau, and the agreement (R²) between the two NDVI products exceeded 0.95 (Figure 5), which indicates that the downscaled NDVI and MODIS NDVI maintain high pixel consistency, further confirming the downscaled product's capability to synchronously capture vegetation dynamics. In contrast, the magnitude of the NDVI variation for the GIMMS data differs significantly from these two products for each grassland type. For some vegetation patches (such as alpine desert steppe, alpine desert, and temperate desert; Figure 5c,e,h), the difference in NDVI between MODIS and the downscaled product ranged from 0.011 to 0.022, with R² ranging from 0.95 to 0.98, while the difference in NDVI between GIMMS and the downscaled product ranged from 0.043 to 0.045, with R² ranging from 0.38 to 0.49. The possible cause of this significant difference may be the nonstationarity of GIMMS NDVI data, leading to uncertainty in detecting vegetation seasonal and interannual variations (Pinzon & Tucker, 2014).

[IMAGE OMITTED. SEE PDF]

Figure 6 shows the linear fitting lines of the downscaled NDVI versus MODIS NDVI for representative grassland types on the Qinghai–Tibet Plateau during the growing season (May–September). The red dot-centered region highlights where the majority of the NDVI values are in each grassland type in the plots. For the downscaled product, there is a noticeable difference in accuracy between the different grassland types. The data accuracy of the alpine meadow type (Figure 6a) is higher (R² = 0.752; RMSE = 0.112; MAE = 0.080) compared to other grassland types and the alpine desert steppe type (Figure 6c) is relatively poorer (R² = 0.588; RMSE = 0.034; MAE = 0.021). The performance of the downscaled product is relatively stable for other grassland types (R²: 0.609–0.734; RMSE: 0.048–0.101; MAE: 0.025–0.083). The results indicate a high agreement between the downscaled NDVI product and the MODIS NDVI product, both in dynamic trends and in spatial patterns.

[IMAGE OMITTED. SEE PDF]

Trend agreement

To assess the capability of the downscaled NDVI product for characterizing vegetation greenness trend changes, the interannual trends of NDVI for the Qinghai–Tibet Plateau based on downscaled NDVI, GIMMS NDVI, and MODIS NDVI using Sen's slope and the Mann–Kendall trend test are shown in Figure 7. Regarding the spatial distribution, the agreement with the MODIS NDVI is higher for the downscaled product. NDVI shows a declining trend in the hinterland of the Qinghai–Tibet Plateau, while lower altitude areas show a significant upward trend (Figure 7d,f). Specifically, the significantly increasing area for the downscaled product accounts for 5.61%, while the decreasing area is 44.12%. For MODIS NDVI, the area of significant increase is 10.25% and the area of decrease is 38.79%. However, GIMMS NDVI shows a declining trend in the southern part of the plateau, covering an area of 29.06%, while lower altitude areas and the western mountainous region show a significant upward trend, covering an area of 17.72% (Figure 7b).

[IMAGE OMITTED. SEE PDF]

DISCUSSION

Development of long-term NDVI products for alpine grassland

As an important indicator of the net primary productivity of vegetation, NDVI is widely used to describe the spatiotemporal dynamics of vegetation activity. For long-term monitoring of vegetation change trends on the Qinghai–Tibet Plateau, previous studies have primarily relied on the GIMMS NDVI data set, which has a wide temporal range but low spatial resolution, and the lack of reliable onboard calibration devices of the AVHRR sensor reduces the accuracy of atmospheric correction, which, to some extent, may lead to significant uncertainties in trend analysis (Nagol et al., 2009). Meanwhile, deficiencies in the data itself may result in contradictory conclusions in studies on vegetation change trends (Cortés et al., 2021; de Jong et al., 2011; Pan et al., 2018). To overcome these limitations, there is an urgent need for NDVI products with long time series and high spatial resolution to more accurately monitor vegetation changes.

In this study, we proposed the NDVI RF downscaled algorithm to generate a long-term time-series NDVI data set, which successfully extended the time series of NDVI data with a long-term duration (1982–2020) and high spatial resolution (250 m), demonstrating high potential in accurately describing the trends of greenness changes and seasonal variation of different vegetation types, especially in the alpine grassland areas of the Qinghai–Tibet Plateau, where it can capture finer details of spatial variations of NDVI.

Uncertainties analysis

The validation phase used MODIS NDVI data as a benchmark, where the inherent accuracy and errors of MODIS data largely affect the precision and accuracy of the downscaled products (Tian et al., 2015). The limited availability of extensive ground-truth data on the Qinghai–Tibet Plateau further constrains the comprehensive validation of these downscaled products, signifying a need for enhanced data collection efforts in future studies. In addition, the application of topographic information in downscaled studies may be another factor contributing to the uncertainty of the product. Topographic variations were considered in early downscaled studies concerning environmental variables (e.g., LST, precipitation), which, as an influential factor influencing the NDVI distribution, has shown great potential in improving the accuracy of the algorithm (Bartkowiak et al., 2019; Xu et al., 2020). However, there is interaction between NDVI and overall hydrothermal conditions (Zhang, He, et al., 2022), and the variation of NDVI with topography on the Qinghai–Tibet Plateau is quite complex (Wang et al., 2020), making it challenging to incorporate topographic data into downscaling studies, where differences may have an impact on NDVI.

The lack of residual corrections and calibration processes in the downscaled results may also introduce uncertainties. Residual correction, accounting for differences between simulated values at coarse resolution and original satellite data (Ulloa et al., 2017), has been demonstrated to enhance downscaled results (Baghanam et al., 2020). For example, residual correction considerably corrects bias obtained by spline interpolation, classification and regression trees, and nearest-neighbor methods (Ashouri et al., 2015; Jia et al., 2011; Xu et al., 2015), because of which it is considered a necessary step in downscaling. However, there are also studies with opposite conclusions, indicating that residual correction may reduce the estimation performance of the model (Duan & Bastiaanssen, 2013; Zhu et al., 2023).

Potential causes of uncertainty

Compared to findings obtained by Ma, Xie et al. (2022), our downscaled NDVI product shows higher accuracy and fewer errors. However, the differences between the two NDVI products are more significant in the alpine region and in some areas to the southeast of the plateau. In alpine vegetation, primarily in the Himalayas, the likely cause of the significant differences in the products is attributed to the inherent limitations in the RF downscaling algorithm. The performance of a machine learning model strongly depends on its performance on the training set, and in this case, DEM contributes the most to the model, which means that the predictive power of the RF downscaled model is greatly affected by DEM. As a result, the algorithm itself may amplify the differences between NDVI products, leading to substantial bias in the downscaled products.

The main reason for the increased bias in the downscaled NDVI product in the southeastern plateau region is the decrease in NDVI sensitivity. Alpine meadow types are densely distributed in the region, NDVI is easily saturated during the growing season (Wunderle et al., 2004), is insensitive to changes in high biomass conditions (Gitelson, 2004; Sakamoto et al., 2010; Zeng et al., 2016), and NDVI no longer continues to increase with vegetation growth, making it difficult to distinguish seasonal variations in vegetation greenness (May et al., 2018). Furthermore, the spatial distribution of NDVI in alpine meadow types is subject to a combination of climatic factors and human activities, and with the instability of global climate change and the increasing intensity of human activities, this has led to highly abrupt spatial changes in grasslands and significant changes in NDVI (Sun, Gong, et al., 2023; Sun, Li, et al., 2023), which may also contribute to the high bias in the downscaled product. On the contrary, the Enhanced Vegetation Index (EVI), also used to monitor vegetation growth, performs well in vegetation types where NDVI saturates, and can clearly reflect the seasonal characteristics of vegetation growth (Bai, 2021; Lin et al., 2008; Ma, Xie, et al., 2022). However, the EVI series has a narrow temporal range, limiting its full potential (Fensholt et al., 2006; Li et al., 2021).

CONCLUSIONS

In this study, based on MODIS and AVHRR vegetation index time-series products, we successfully developed a suite of decadal-scale vegetation index data set with long-term duration (1982–2020) and high spatial resolution (250 m), which provides strong support for the long-term dynamic monitoring of alpine grassland in the Qinghai–Tibet Plateau.

The RF downscaling algorithm demonstrates excellent accuracy and high agreement with MODIS NDVI. In most regions of the Qinghai–Tibet Plateau, the RMSE ranges from 0 to 0.075, MAE ranges from 0 to 0.05, and R² is typically higher than 0.7, which indicates that the algorithm can accurately simulate the spatial variation of vegetation indices, providing a reliable data foundation for additional surface ecological studies.

AUTHOR CONTRIBUTIONS

Xiali Yang: Data curation; writing—original draft. Xiaodong Huang: Conceptualization; funding acquisition; methodology; supervision; writing—review and editing. Ying Ma: Validation. Yuxin Li: Data curation; visualization. Qisheng Feng: Data curation; methodology. Tiangang Liang: Methodology; supervision; writing—review and editing.

ACKNOWLEDGMENTS

This work was supported by the Natural Science Foundation of China Projects (41971293).

CONFLICT OF INTEREST STATEMENT

The authors declare no conflicts of interest.

DATA AVAILABILITY STATEMENT

The data that support the findings of this study are available from the corresponding author upon reasonable request.

References

Ali, S., Liu, D., Fu, Q., Cheema, M. J. M., Pham, Q. B., Rahaman, M. M., Dang, T. D., & Anh, D. T. (2021). Improving the resolution of GRACE data for spatio‐temporal groundwater storage assessment. Remote Sensing, 13(17), 3513. https://doi.org/10.3390/rs13173513

An, S., Zhang, X., Chen, X., Yan, D., & Henebry, G. (2018). An exploration of terrain effects on land surface phenology across the Qinghai–Tibet plateau using Landsat ETM+ and OLI data. Remote Sensing, 10(7), 1069. https://doi.org/10.3390/rs10071069

An, Y., Gao, W., Gao, Z., Liu, C., & Shi, R. (2015). Trend analysis for evaluating the consistency of Terra MODIS and SPOT VGT NDVI time series products in China. Frontiers of Earth Science, 9(1), 125–136. https://doi.org/10.1007/s11707-014-0428-9

Anderson, M. C., Yang, Y., Xue, J., Knipper, K. R., Yang, Y., Gao, F., Hain, C. R., Kustas, W. P., Cawse‐Nicholson, K., Hulley, G., Fisher, J. B., Alfieri, J. G., Meyers, T. P., Prueger, J., Baldocchi, D. D., & Rey‐Sanchez, C. (2021). Interoperability of ECOSTRESS and Landsat for mapping evapotranspiration time series at sub‐field scales. Remote Sensing of Environment, 252, 112189. https://doi.org/10.1016/j.rse.2020.112189

Ashouri, H., Hsu, K.‐L., Sorooshian, S., Braithwaite, D. K., Knapp, K. R., Cecil, L. D., Nelson, B. R., & Prat, O. P. (2015). PERSIANN‐CDR: Daily precipitation climate data record from multisatellite observations for hydrological and climate studies. Bulletin of the American Meteorological Society, 96(1), 69–83. https://doi.org/10.1175/bams-d-13-00068.1

Baghanam, A. H., Eslahi, M., Sheikhbabaei, A., & Seifi, A. J. (2020). Assessing the impact of climate change over the northwest of Iran: An overview of statistical downscaling methods. Theoretical and Applied Climatology, 141(3–4), 1135–1150. https://doi.org/10.1007/s00704-020-03271-8

Bai, Y. (2021). Analysis of vegetation dynamics in the Qinling–Daba Mountains region from MODIS time series data. Ecological Indicators, 129, 108029. https://doi.org/10.1016/j.ecolind.2021.108029

Bartkowiak, P., Castelli, M., & Notarnicola, C. (2019). Downscaling land surface temperature from MODIS dataset with random forest approach over alpine vegetated areas. Remote Sensing, 11(11), 1319. https://doi.org/10.3390/rs11111319

Cao, C., Weinreb, M., & Xu, H. (2004). Predicting simultaneous nadir overpasses among polar‐orbiting meteorological satellites for the intersatellite calibration of radiometers. Journal of Applied Remote Sensing, 21(4), 537–542. https://doi.org/10.1175/1520-0426(2004)021<0537:PSNOAP>2.0.CO;2

Cao, R., Chen, Y., Chen, J., Zhu, X., & Shen, M. (2020). Thick cloud removal in Landsat images based on autoregression of Landsat time‐series data. Remote Sensing of Environment, 249, 112001. https://doi.org/10.1016/j.rse.2020.112001

Chen, F., Liu, Z., Zhong, H., & Wang, S. (2021). Exploring the applicability and scaling effects of satellite‐observed spring and autumn phenology in complex terrain regions using four different spatial resolution products. Remote Sensing, 13(22), 4582. https://doi.org/10.3390/rs13224582

Cheng, J., Dai, Y., Yuan, Y., & Zhu, H. (2020). A simple analysis of multimodal data fusion. In G. Wang, R. Ko, M. Bhuiyan, & Y. Pan (Eds.), 2020 IEEE 19th international conference on trust, security and privacy in computing and communications (pp. 1472–1475). IEEE. https://doi.org/10.1109/TrustCom50675.2020.00199

Cong, N., Wang, T., Nan, H., Ma, Y., Wang, X., Myneni, R. B., & Piao, S. (2013). Changes in satellite‐derived spring vegetation green‐up date and its linkage to climate in China from 1982 to 2010: A multimethod analysis. Global Change Biology, 19(3), 881–891. https://doi.org/10.1111/gcb.12077

Cortés, J., Mahecha, M., Reichstein, M., Myneni, R., Chen, C., & Brenning, A. (2021). Where are global vegetation greening and browning trends significant? Global Change Biology, 48(6), e2020GL091496. https://doi.org/10.1029/2020GL091496

Cui, J., Zhang, X., & Luo, M. (2018). Combining linear pixel unmixing and STARFM for spatiotemporal fusion of gaofen‐1 wide field of view imagery and MODIS imagery. Remote Sensing, 10(7), 1047. https://doi.org/10.3390/rs10071047

Cui, L., & Shi, J. (2010). Temporal and spatial response of vegetation NDVI to temperature and precipitation in eastern China. Journal of Hydrology, 20(2), 163–176. https://doi.org/10.1007/s11442-010-0163-4

Ding, B., Feng, L., Ba, S., Jiang, X., Liu, G., & Liu, W. (2023). Temperature drives elevational diversity patterns of different types of organisms in Qinghai–Tibetan Plateau wetlands. iScience, 26(8), 107252. https://doi.org/10.1016/j.isci.2023.107252

Ding, M., Li, L., Zhang, Y., Sun, X., Liu, L., Gao, J., Wang, Z., & Li, Y. (2015). Start of vegetation growing season on the Tibetan Plateau inferred from multiple methods based on GIMMS and SPOT NDVI data. Journal of Hydrology, 25(2), 131–148. https://doi.org/10.1007/s11442-015-1158-y

Ding, M., Zhang, Y., Liu, L., Zhang, W., Wang, Z., & Bai, W. (2007). The relationship between NDVI and precipitation on the Tibetan Plateau. Journal of Applied Remote Sensing, 17(3), 259–268. https://doi.org/10.1007/s11442-007-0259-7

Du, J. Q., Shu, J. M., Wang, Y. H., Li, Y. C., Zhang, L. B., & Guo, Y. (2014). Comparison of GIMMS and MODIS normalized vegetation index composite data for Qinghai‐Tibet Plateau. Chinese Journal of Applied Ecology, 25(2), 533–544. https://doi.org/10.13287/j.1001-9332.2014.0056

Duan, Z., & Bastiaanssen, W. G. M. (2013). First results from Version 7 TRMM 3B43 precipitation product in combination with a new downscaling‐calibration procedure. Remote Sensing of Environment, 131, 1–13. https://doi.org/10.1016/j.rse.2012.12.002

Faisal, B., Rahman, H., Sharifee, N., Sultana, N., Islam, M., & Ahammad, T. (2019). Remotely sensed boro rice production forecasting using MODIS‐NDVI: A Bangladesh perspective. Applied Mechanics and Materials, 1(3), 356–375. https://doi.org/10.3390/agriengineering1030027

Fensholt, R., Langanke, T., Rasmussen, K., Reenberg, A., Prince, S. D., Tucker, C., Scholes, R. J., Le, Q. B., Bondeau, A., Eastman, R., Epstein, H., Gaughan, A. E., Hellden, U., Mbow, C., Olsson, L., Paruelo, J., Schweitzer, C., Seaquist, J., & Wessels, K. (2012). Greenness in semi‐arid areas across the globe 1981–2007—An Earth Observing Satellite based analysis of trends and drivers. Remote Sensing of Environment, 121, 144–158. https://doi.org/10.1016/j.rse.2012.01.017

Fensholt, R., & Proud, S. R. (2012). Evaluation of Earth Observation based global long term vegetation trends—Comparing GIMMS and MODIS global NDVI time series. Remote Sensing of Environment, 119, 131–147. https://doi.org/10.1016/j.rse.2011.12.015

Fensholt, R., Sandholt, I., & Stisen, S. (2006). Evaluating MODIS, MERIS, and VEGETATION: Vegetation indices using in situ measurements in a semiarid environment. IEEE Transactions on Geoscience and Remote Sensing, 44(7), 1774–1786. https://doi.org/10.1109/TGRS.2006.875940

Ge, Y., Jin, Y., Stein, A., Chen, Y., Wang, J., Wang, J., Cheng, Q., Bai, H., Liu, M., & Atkinson, P. (2019). Principles and methods of scaling geospatial Earth science data. Environmental Earth Sciences, 197, 102897. https://doi.org/10.1016/j.earscirev.2019.102897

Gitelson, A. A. (2004). Wide dynamic range vegetation index for remote quantification of biophysical characteristics of vegetation. Journal of Plant Physiology, 161(2), 165–173. https://doi.org/10.1078/0176-1617-01176

Gao, F., Masek, J., Schwaller, M., & Hall, F. (2006). On the blending of the landsat and MODIS surface reflectance: Predicting daily landsat surface reflectance. IEEE Transactions on Geoscience and Remote Sensing, 44(8), 2207–2218. https://doi.org/10.1109/TGRS.2006.872081

Huang, S., Zhang, X., Wang, C., & Chen, N. (2023). Two‐step fusion method for generating 1 km seamless multi‐layer soil moisture with high accuracy in the Qinghai–Tibet plateau. International Journal of Remote Sensing, 197, 346–363. https://doi.org/10.1016/j.isprsjprs.2023.02.009

Huete, A., Didan, K., Miura, T., Rodriguez, E. P., Gao, X., & Ferreira, L. G. (2002). Overview of the radiometric and biophysical performance of the MODIS vegetation indices. Remote Sensing of Environment, 83(1–2), 195–213. https://doi.org/10.1016/S0034-4257(02)00096-2

Jia, S., Zhu, W., Lu, A., & Yan, T. (2011). A statistical spatial downscaling algorithm of TRMM precipitation based on NDVI and DEM in the Qaidam Basin of China. Remote Sensing of Environment, 115(12), 3069–3079. https://doi.org/10.1016/j.rse.2011.06.009

Jiang, L., Tarpley, J. D., Mitchell, K. E., Zhou, S., Kogan, F. N., & Guo, W. (2008). Adjusting for long‐term anomalous trends in NOAA's global vegetation index data sets. IEEE Transactions on Geoscience and Remote Sensing, 46(2), 409–422. https://doi.org/10.1109/TGRS.2007.902844

Jiang, Z., Huete, A. R., Chen, J., Chen, Y., Li, J., Yan, G., & Zhang, X. (2006). Analysis of NDVI and scaled difference vegetation index retrievals of vegetation fraction. Remote Sensing of Environment, 101(3), 366–378. https://doi.org/10.1016/j.rse.2006.01.003

de Jong, R., de Bruin, S., de Wit, A., Schaepman, M. E., & Dent, D. L. (2011). Analysis of monotonic greening and browning trends from global NDVI time‐series. Remote Sensing of Environment, 115(2), 692–702. https://doi.org/10.1016/j.rse.2010.10.011

Ju, J., & Roy, D. P. (2008). The availability of cloud‐free Landsat ETM+ data over the conterminous United States and globally. Remote Sensing of Environment, 112(3), 1196–1211. https://doi.org/10.1016/j.rse.2007.08.011

Karbalaye Ghorbanpour, A., Hessels, T., Moghim, S., & Afshar, A. (2021). Comparison and assessment of spatial downscaling methods for enhancing the accuracy of satellite‐based precipitation over Lake Urmia Basin. Journal of Hydrology, 596, 126055. https://doi.org/10.1016/j.jhydrol.2021.126055

Ke, Y., Im, J., Park, S., & Gong, H. (2016). Downscaling of MODIS one kilometer evapotranspiration using Landsat‐8 data and machine learning approaches. Remote Sensing, 8(3), 215. https://doi.org/10.3390/rs8030215

Kerdiles, H., & Grondona, M. O. (1995). NOAA‐AVHRR NDVI decomposition and subpixel classification using linear mixing in the Argentinean Pampa. International Journal of Remote Sensing, 16(7), 1303–1325. https://doi.org/10.1080/01431169508954478

Lan, S., & Dong, Z. (2022). Incorporating vegetation type transformation with NDVI time‐series to study the vegetation dynamics in Xinjiang. Science of the Total Environment, 14(1), 582. https://doi.org/10.3390/su14010582

Landman, W. (2010). Climate change 2007: The physical science basis. Science of the Total Environment, 92(1), 86–87. https://doi.org/10.1080/03736245.2010.480842

van Leeuwen, W. J. D., Orr, B. J., Marsh, S. E., & Herrmann, S. M. (2006). Multi‐sensor NDVI data continuity: Uncertainties and implications for vegetation monitoring applications. Remote Sensing of Environment, 100(1), 67–81. https://doi.org/10.1016/j.rse.2005.10.002

Li, D., Wu, S., Liu, L., Zhang, Y., & Li, S. (2018). Vulnerability of the global terrestrial ecosystems to climate change. Global Change Biology, 24(9), 4095–4106. https://doi.org/10.1111/gcb.14327

Li, M., Cao, S., Zhu, Z., Wang, Z., Myneni, R. B., & Piao, S. (2023). Spatiotemporally consistent global dataset of the GIMMS Normalized Difference Vegetation Index (PKU GIMMS NDVI) from 1982 to 2022. Earth System Science Data, 15(9), 4181–4203. https://doi.org/10.5194/essd-15-4181-2023

Li, P., Hu, Z., & Liu, Y. (2020). Shift in the trend of browning in Southwestern Tibetan Plateau in the past two decades. Agricultural and Forest Meteorology, 287, 107950. https://doi.org/10.1016/j.agrformet.2020.107950

Li, Y., Jiao, Z., Zhao, K., Dong, Y., Zhou, Y., Zeng, Y., Xu, H., Zhang, X., Hu, T., & Cui, L. (2021). Influence of varying solar zenith angles on land surface phenology derived from vegetation indices: A case study in the Harvard Forest. Remote Sensing, 13(20), 4126. https://doi.org/10.3390/rs13204126

Lima, C. H. R., Kwon, H.‐H., & Kim, Y.‐T. (2021). A Bayesian Kriging model applied for spatial downscaling of daily rainfall from GCMs. Journal of Hydrology, 597, 126095. https://doi.org/10.1016/j.jhydrol.2021.126095

Lin, W., Shi, R., & Shi, J. (2008). Compared MODIS‐NDVI with MODIS‐EVI in forecast crop yield. In G. Li, Z. Jia, & Z. Fu (Eds.), 2008 proceedings of information technology and environmental system sciences (pp. 212–216). WOS.

Liu, X., & Chen, B. (2000). Climatic warming in the Tibetan Plateau during recent decades. International Journal of Climatology, 20(14), 1729–1742. https://doi.org/10.1002/1097-0088(20001130)20:14<1729::AID-JOC556>3.0.CO;2-Y

Liu, Y., Jing, W., Wang, Q., & Xia, X. (2020). Generating high‐resolution daily soil moisture by using spatial downscaling techniques: A comparison of six machine learning algorithms. Agricultural and Forest Meteorology, 141, 103601. https://doi.org/10.1016/j.advwatres.2020.103601

Liu, Y., Li, Z., Chen, Y., Li, Y., Li, H., Xia, Q., & Kayumba, P. M. (2022). Evaluation of consistency among three NDVI products applied to High Mountain Asia in 2000–2015. Remote Sensing of Environment, 269, 112821. https://doi.org/10.1016/j.rse.2021.112821

Lv, J., Zhao, W., Hua, T., Zhang, L., & Pereira, P. (2023). Multiple Greenness Indexes revealed the vegetation greening during the growing season and winter on the Tibetan Plateau despite regional variations. Remote Sensing, 15(24), 5697. https://doi.org/10.3390/rs15245697

Ma, C., Xie, Y., Duan, S., Qin, W., Guo, Z., Xi, G., Zhang, X., Bie, Q., Duan, H., & He, L. (2022). Characterization of spatio‐temporal patterns of grassland utilization intensity in the Selinco watershed of the Qinghai–Tibetan Plateau from 2001 to 2019 based on multisource remote sensing and artificial intelligence algorithms. GIScience & Remote Sensing, 59(1), 2217–2246. https://doi.org/10.1080/15481603.2022.2153447

Ma, Z., Dong, C., Lin, K., Yan, Y., Luo, J., Jiang, D., & Chen, X. (2022). A global 250‐m downscaled NDVI product from 1982 to 2018. Remote Sensing, 14(15), 3639. https://doi.org/10.3390/rs14153639

Marshall, M., Okuto, E., Kang, Y., Opiyo, E., & Ahmed, M. (2016). Global assessment of Vegetation Index and Phenology Lab (VIP) and Global Inventory Modeling and Mapping Studies (GIMMS) version 3 products. Bulletin of the American Meteorological Society, 13(3), 625–639. https://doi.org/10.5194/bg-13-625-2016

Martinez, A. I., & Labib, S. M. (2023). Demystifying Normalized Difference Vegetation Index (NDVI) for greenness exposure assessments and policy interventions in urban greening. Environmental Research, 220, 115155. https://doi.org/10.1016/j.envres.2022.115155

May, J. L., Parker, T., Unger, S., & Oberbauer, S. F. (2018). Short term changes in moisture content drive strong changes in Normalized Difference Vegetation Index and gross primary productivity in four Arctic moss communities. Remote Sensing of Environment, 212, 114–120. https://doi.org/10.1016/j.rse.2018.04.041

Moreno‐Martinez, A., Moneta, M., Valls, G. C., Martino, L., Robinson, N., Allred, B., & Running, S. W. (2018). Interpolation and gap filling of landsat reflectance time series. In IGARSS 2018 ‐ 2018 IEEE international geoscience and remote sensing symposium. https://doi.org/10.1109/igarss.2018.8517503

Nagol, J., Vermote, E., & Prince, S. (2014). Quantification of impact of orbital drift on inter‐annual trends in AVHRR NDVI Data. Remote Sensing, 6(7), 6680–6687. https://doi.org/10.3390/rs6076680

Nagol, J. R., Vermote, E. F., & Prince, S. D. (2009). Effects of atmospheric variation on AVHRR NDVI data. Remote Sensing of Environment, 113(2), 392–397. https://doi.org/10.1016/j.rse.2008.10.007

Pan, N., Feng, X., Fu, B., Wang, S., Ji, F., & Pan, S. (2018). Increasing global vegetation browning hidden in overall vegetation greening: Insights from time‐varying trends. Remote Sensing of Environment, 214, 59–72. https://doi.org/10.1016/j.rse.2018.05.018

Pinzon, J., & Tucker, C. (2014). A non‐stationary 1981–2012 AVHRR NDVI3g time series. Remote Sensing, 6(8), 6929–6960. https://doi.org/10.3390/rs6086929

Qu, Y., Zhu, Z., Montzka, C., Chai, L., Liu, S., Ge, Y., Liu, J., Lu, Z., He, X., Zheng, J., & Han, T. (2021). Inter‐comparison of several soil moisture downscaling methods over the Qinghai–Tibet Plateau, China. Journal of Hydrology, 592, 125616. https://doi.org/10.1016/j.jhydrol.2020.125616

Rodríguez, E., Morris, C. S., & Belz, J. E. (2006). A global assessment of the SRTM performance. Photogrammetric Engineering & Remote Sensing, 72(3), 249–260. https://doi.org/10.14358/pers.72.3.249

Sakamoto, T., Wardlow, B. D., Gitelson, A. A., Verma, S. B., Suyker, A. E., & Arkebauer, T. J. (2010). A two‐step filtering approach for detecting maize and soybean phenology with time‐series MODIS data. Remote Sensing of Environment, 114(10), 2146–2159. https://doi.org/10.1016/j.rse.2010.04.019

Sano, E. E., Ferreira, L. G., Asner, G. P., & Steinke, E. T. (2007). Spatial and temporal probabilities of obtaining cloud‐free Landsat images over the Brazilian tropical savanna. International Journal of Remote Sensing, 28(12), 2739–2752. https://doi.org/10.1080/01431160600981517

Schneider, M., & Reinartz, P. (2009). Matching of high resolution optical data to a shaded DEM. In 2009 IEEE international geoscience and remote sensing symposium. https://doi.org/10.1109/igarss.2009.5417740

Sdraka, M., Papoutsis, I., Psomas, B., Vlachos, K., Ioannidis, K., Karantzalos, K., Gialampoukidis, I., & Vrochidis, S. (2022). Deep learning for downscaling remote sensing images: Fusion and super‐resolution. IEEE Geoscience and Remote Sensing Magazine, 10(3), 202–255. https://doi.org/10.1109/MGRS.2022.3171836

Shen, H., Li, X., Cheng, Q., Zeng, C., Yang, G., Li, H., & Zhang, L. (2015). Missing information reconstruction of remote sensing data: A technical review. IEEE Geoscience and Remote Sensing Magazine, 3(3), 61–85. https://doi.org/10.1109/MGRS.2015.2441912

Shen, M., Piao, S., Jeong, S. J., Zhou, L., Zeng, Z., Ciais, P., Chen, D., Huang, M., Jin, C. S., Li, L. Z., Li, Y., Myneni, R. B., Yang, K., Zhang, G., Zhang, Y., & Yao, T. (2015). Evaporative cooling over the Tibetan Plateau induced by vegetation growth. Proceedings of the National Academy of Sciences of the United States of America, 112(30), 9299–9304. https://doi.org/10.1073/pnas.1504418112

Somers, B., Asner, G. P., Tits, L., & Coppin, P. (2011). Endmember variability in spectral mixture analysis: A review. Remote Sensing of Environment, 115(7), 1603–1616. https://doi.org/10.1016/j.rse.2011.03.003

Sun, L., Li, H., Wang, J., Chen, Y., Xiong, N., Wang, Z., Wang, J., & Xu, J. (2023). Impacts of climate change and human activities on NDVI in the Qinghai–Tibet plateau. Remote Sensing, 15(3), 587. https://doi.org/10.3390/rs15030587

Sun, M., Gong, A., Zhao, X., Liu, N., Si, L., & Zhao, S. (2023). Reconstruction of a monthly 1 km NDVI time series product in China using random forest methodology. Remote Sensing, 15(13), 3353. https://doi.org/10.3390/rs15133353

Sun, Z., Ouyang, X., Li, H., & Wang, J. (2024). A deep learning‐based spatio‐temporal NDVI data fusion model. Journal of Resources and Ecology, 15(1), 214–226. https://doi.org/10.5814/j.issn.1674-764x.2024.01.019

Sun, Z., & Wang, J. (2022). The 30 m‐NDVI‐based alpine grassland changes and climate impacts in the Three‐River Headwaters Region on the Qinghai–Tibet Plateau from 1990 to 2018. Journal of Resources and Ecology, 13(2), 186–195. https://doi.org/10.5814/j.issn.1674-764x.2022.02.002

Tian, F., Fensholt, R., Verbesselt, J., Grogan, K., Horion, S., & Wang, Y. (2015). Evaluating temporal consistency of long‐term global NDVI datasets for trend analysis. Remote Sensing of Environment, 163, 326–340. https://doi.org/10.1016/j.rse.2015.03.031

Tucker, C. J. (1979). Red and photographic infrared linear combinations for monitoring vegetation. Remote Sensing of Environment, 8(2), 127–150. https://doi.org/10.1016/0034-4257(79)90013-0

Tucker, C. J., Pinzon, J. E., Brown, M. E., Slayback, D. A., Pak, E. W., Mahoney, R., Vermote, E. F., & El Saleous, N. (2005). An extended AVHRR 8‐km NDVI dataset compatible with MODIS and SPOT vegetation NDVI data. International Journal of Remote Sensing, 26(20), 4485–4498. https://doi.org/10.1080/01431160500168686

Ulloa, J., Ballari, D., Campozano, L., & Samaniego, E. (2017). Two‐step downscaling of TRMM 3B43 V7 precipitation in contrasting climatic regions with sparse monitoring: The case of Ecuador in Tropical South America. Remote Sensing, 9(7), 758. https://doi.org/10.3390/rs9070758

Wang, T., Yang, M., Yan, S., Geng, G., Li, Q., & Wang, F. (2020). Temporal and spatial vegetation index variability and response to temperature and precipitation in the Qinghai–Tibet plateau using GIMMS NDVI. Photogrammetric Engineering and Remote Sensing, 29(6), 4385–4395. https://doi.org/10.15244/pjoes/120768

Wang, Y., Lv, W., Xue, K., Wang, S., Zhang, L., Hu, R., Zeng, H., Xu, X., Li, Y., Jiang, L., Hao, Y., Du, J., Sun, J., Dorji, T., Piao, S., Wang, C., Luo, C., Zhang, Z., Chang, X., … Niu, H. (2022). Grassland changes and adaptive management on the Qinghai–Tibetan Plateau. Nature Reviews Earth & Environment, 3(10), 668–683. https://doi.org/10.1038/s43017-022-00330-8

Warren, R., Price, J., Fischlin, A., de la Nava Santos, S., & Midgley, G. (2011). Increasing impacts of climate change upon ecosystems with increasing global mean temperature rise. Climatic Change, 106(2), 141–177. https://doi.org/10.1007/s10584-010-9923-5

Wu, S., Liu, L., Gao, J., & Wang, W. (2019). Integrate risk from climate change in China under global warming of 1.5 and 2.0°C. Earth's Future, 7(12), 1307–1322. https://doi.org/10.1029/2019EF001194

Wu, S., Yin, Y., Zheng, D., & Yang, Q. (2007). Climatic trends over the Tibetan Plateau during 1971–2000. Journal of Applied Remote Sensing, 17(2), 141–151. https://doi.org/10.1007/s11442-007-0141-7

Wunderle, S., Oesch, D., Hauser, A., & Foppa, N. (2004). Operational estimation of vegetation index (NDVI), vegetation cover and leaf area index using NOAA‐AVHRR data in an alpine environment. In M. Owe, G. D'Urso, J. Moreno, & A. Calera (Eds.), Proceedings of SPIE (Vol. 5232, pp. 20–29). Remote Sensing for Agriculture, Ecosystems and Hydrology V. https://doi.org/10.1117/12.511020

Xu, J., Zhang, F., Jiang, H., Hu, H., Zhong, K., Jing, W., Yang, J., & Jia, B. (2020). Downscaling Aster land surface temperature over urban areas with machine learning‐based area‐to‐point regression Kriging. Remote Sensing, 12, 1082. https://doi.org/10.3390/rs12071082

Xu, S., Wu, C., Wang, L., Gonsamo, A., Shen, Y., & Niu, Z. (2015). A new satellite‐based monthly precipitation downscaling algorithm with non‐stationary relationship between precipitation and land surface characteristics. Remote Sensing of Environment, 162, 119–140. https://doi.org/10.1016/j.rse.2015.02.024

Yan, X., Chen, H., Tian, B., Sheng, S., Wang, J., & Kim, J. S. (2021). A Downscaling‐merging scheme for improving daily spatial precipitation estimates based on Random Forest and Cokriging. Remote Sensing, 13(11), 2040. https://doi.org/10.3390/rs13112040

Yao, T., Thompson, L. G., Mosbrugger, V., Zhang, F., Ma, Y., Luo, T., Xu, B., Yang, X., Joswiak, D. R., Wang, W., Joswiak, M. E., Devkota, L. P., Tayal, S., Jilani, R., & Fayziev, R. (2012). Third Pole Environment (TPE). Environmental Development, 3, 52–64. https://doi.org/10.1016/j.envdev.2012.04.002

You, Q., Min, J., & Kang, S. (2016). Rapid warming in the Tibetan Plateau from observations and CMIP5 models in recent decades. International Journal of Climatology, 36(6), 2660–2670. https://doi.org/10.1002/joc.4520

Yu, H., Guo, J., Cheng, Y., & Lou, Q. (2013). Techniques and methods of spatial data fusion. In J. Zhang, Z. Wang, S. Zhu, & X. Meng (Eds.), Applied mechanics and materials (Vol. 263–266, pp. 3274–3278). Information Technology Applications in Industry, PTS 1‐4. https://doi.org/10.4028/www.scientific.net/AMM.263-266.3274

Yuan, H., Matthew, C., He, X. Z., Sun, Y., Liu, Y., Zhang, T., Gao, X., Yan, C., Chang, S., & Hou, F. (2022). Seasonal variation in soil and herbage CO2 efflux for a sheep‐grazed alpine meadow on the north‐east Qinghai–Tibetan plateau and estimated net annual CO2 exchange. Frontiers in Plant Science, 13, 860739. https://doi.org/10.3389/fpls.2022.860739

Yuan, W., Zheng, Y., Piao, S., Ciais, P., Lombardozzi, D., Wang, Y., Ryu, Y., Chen, G., Dong, W., Hu, Z., Jain, A. K., Jiang, C., Kato, E., Li, S., Lienert, S., Liu, S., Nabel, J., Qin, Z., Quine, T., … Yang, S. (2019). Increased atmospheric vapor pressure deficit reduces global vegetation growth. Science Advances, 5(8), 1396. https://doi.org/10.1126/sciadv.aax1396

Zeng, L., Wardlow, B. D., Wang, R., Shan, J., Tadesse, T., Hayes, M. J., & Li, D. (2016). A hybrid approach for detecting corn and soybean phenology with time‐series MODIS data. Remote Sensing of Environment, 181, 237–250. https://doi.org/10.1016/j.rse.2016.03.039

Zhang, G., Zhang, Y., Dong, J., & Xiao, X. (2013). Green‐up dates in the Tibetan Plateau have continuously advanced from 1982 to 2011. Proceedings of the National Academy of Sciences of the United States of America, 110(11), 4309–4314. https://doi.org/10.1073/pnas.1210423110

Zhang, Q., Cheng, J., & Wang, N. (2022). Fusion of all‐weather land surface temperature from AMSR‐E and MODIS data using Random Forest regression. IEEE Geoscience and Remote Sensing Letters, 19, 2502705. https://doi.org/10.1109/LGRS.2021.3120431

Zhang, Y., He, Y., Li, Y., & Jia, L. (2022). Spatiotemporal variation and driving forces of NDVI from 1982 to 2015 in the Qinba Mountains, China. Environmental Science and Pollution Research, 29(34), 52277–52288. https://doi.org/10.1007/s11356-022-19502-6

Zhong, L., Ma, Y., Xue, Y., & Piao, S. (2019). Climate change trends and impacts on vegetation greening over the Tibetan Plateau. Journal of Applied Remote Sensing, 124(14), 7540–7552. https://doi.org/10.1029/2019JD030481

Zhou, D., Fan, G., Huang, R., Fang, Z., Liu, Y., & Li, H. (2007). Interannual variability of the normalized difference vegetation index on the Tibetan plateau and its relationship with climate change. Applied Mechanics and Materials, 24(3), 474–484. https://doi.org/10.1007/s00376-007-0474-2

Zhou, J., Jia, L., & Menenti, M. (2015). Reconstruction of global MODIS NDVI time series: Performance of Harmonic ANalysis of Time Series (HANTS. Remote Sensing of Environment, 163, 217–228. https://doi.org/10.1016/j.rse.2015.03.018

Zhu, H., Liu, H., Zhou, Q., & Cui, A. (2023). Towards an accurate and reliable downscaling scheme for high‐spatial‐resolution precipitation data. Remote Sensing, 15(10), 2640. https://doi.org/10.3390/rs15102640

Zurita‐Milla, R., Kaiser, G., Clevers, J. G. P. W., Schneider, W., & Schaepman, M. E. (2009). Downscaling time series of MERIS full resolution data to monitor vegetation seasonal dynamics. Remote Sensing of Environment, 113(9), 1874–1885. https://doi.org/10.1016/j.rse.2009.04.011