Decadal trends in global grassland growth peaks and their drivers since the 1980s

You, Cuihai; Chen, Shiping; Tu, Zhiqin; Bian, Chenyu; Cui, Erqian; Huang, Kun; Wan, Fangxiu; Ping, Jiaye; Wei, Ning; Xia, Jianyang

doi:10.1038/s41467-025-64565-x

Download PDF

Article
Open access
Published: 28 October 2025

Decadal trends in global grassland growth peaks and their drivers since the 1980s

Nature Communications volume 16, Article number: 9501 (2025) Cite this article

6970 Accesses
1 Citations
11 Altmetric
Metrics details

Subjects

Abstract

Grasslands, Earth’s most widespread terrestrial ecosystems, are vital for global carbon sequestration and food security. A key indicator of these functions is vegetation growth peak, reflecting maximum seasonal productivity. However, whether this growth peak consistently increases across global grasslands remains unknown. Here, using satellite-derived vegetation data from 1982 to 2021, we reveal a widespread increase in grassland growth peaks, with 71% (20/28) IPCC climate regions exhibiting significant trends. However, this trend reverses between 1998 and 2009 in 64% of regions, notably across the Tibetan Plateau and East Asia. The reversal is supported by a global gross primary productivity dataset generated via machine learning, trained on eddy covariance flux data. The interruption is primarily associated with a global-scale decadal drought. These findings demonstrate that persistent drought can disrupt the upward trajectory of grassland growth peaks, posing a spatially uneven but widespread threat to ecological functions of grasslands under climate change.

Introduction

Grasslands, the world’s most widespread terrestrial ecosystems, cover approximately 40% of the Earth’s surface and support nearly 38% of the global human population¹. Beyond their vital role in maintaining biodiversity and providing resources for livestock, grasslands have emerged as crucial players in regulating global climate change². Recent studies underscore their importance in carbon sequestration, showing that grasslands significantly contribute to interannual variability and long-term trends in terrestrial carbon sinks^3,4,5. A key indicator of these ecological functions is the peak in vegetation growth, referring to the seasonal maximum in vegetation photosynthetic activity, greenness, or aboveground biomass, with the specific meaning varying depending on the metric used, and serving as a critical indicator of the terrestrial ecosystem’s productivity^6,7. Understanding how grassland growth peaks vary across space and time is critical for predicting the future resilience of terrestrial ecosystem functions under changing climate conditions.

Since the 1980s, global vegetation growth peaks have generally increased, driven by CO₂ fertilization, nitrogen deposition, and land-use changes, including cropland expansion^6,8,9. However, this trend is not uniform, with substantial variations among grassland regions. For instance, while the Northwestern Great Plains, Mongolian Plateau, and African Savannas show long-term growth peak increases linked to higher precipitation^10,11,12, other regions exhibit divergent trends. In Asian drylands, growth peaks have shown a weak decline since the 1990s, contrasting sharply with upward trends observed elsewhere¹³. The Pampas region has seen little to no significant change in vegetation growth peaks from the 1990s to the 2010s¹⁴. Vegetation growth peaks in China’s arid grasslands have experienced periodic fluctuations, including a notable decline from 1996 to 2005 despite an overall increase from 1982 to 2015¹⁵. A recent study suggests that Asian grasslands are more sensitive to extreme drought than North American grasslands, likely due to differences in baseline climate conditions and ecosystem adaptation strategies¹⁶. These discrepancies underscore the need for a deeper investigation into the drivers behind such regional differences, especially in the context of global climate change.

As predominantly water-limited ecosystems, grasslands are highly sensitive to drought conditions^1,17. Drought represents an abnormal period of low water availability and thus can significantly disrupt ecosystem function over annual or longer timescales¹⁸. Model projections suggest that future droughts will become more frequent, severe, and longer lasting than those experienced in recent decades^19,20. Periodic prolonged droughts in regions like North America and China have already caused substantial declines in vegetation growth and carbon uptake^21,22,23. Furthermore, a recent study indicates that the frequency and impact of prolonged droughts have increased globally since the 1980s, with grasslands being the most affected, experiencing the greatest declines in vegetation growth²⁴. Therefore, it is critical to understand whether grasslands can maintain the observed upward trend in growth peaks or if increasing drought intensity will threaten their ecological functions by disrupting the growth peak increases.

Here, we examine whether grassland growth peaks have persistently increased over the past four decades. We analyze maximum Normalized Difference Vegetation Index (NDVI_max) from the Global Inventory Modeling and Mapping Studies (GIMMS) dataset (1982–2021)²⁵, and identify temporal disruptions using the breaks for additive season and trend (BFAST) method²⁶. To enhance our analysis, we generate a globally gross primary productivity (GPP) dataset by upscaling eddy covariance flux measurements using machine learning. Additionally, we validate NDVI-derived patterns using independent vegetation indices, including eddy covariance light use efficiency GPP (EC-LUE GPP)²⁷, the enhanced vegetation index (EVI) from MODIS, and solar-induced chlorophyll fluorescence (SIF) data²⁸. Growth peaks were quantified using the maximum values of these key indicators (NDVI_max, EVI_max, SIF_max, and GPP_max), which reflect different aspects of vegetation function, such as canopy greenness (NDVI, EVI), photosynthetic activity (GPP), and chlorophyll fluorescence (SIF). Based on these multiple lines of evidence, we examine regional variations across 28 IPCC-defined subregions and show that widespread drought, marked by sharp declines in precipitation, is the primary driver in reversing the upward trajectory of grassland growth peaks.

Results

Interruptions in the increasing trend of grassland growth peaks during 1998–2009

Over the past four decades (1982 to 2021), global grassland NDVI_max exhibited a significant linear increase at a rate of 0.55 ± 0.08 × 10⁻³ yr⁻¹ (t (38) = 7.06, P < 0.001, 95% CI [0.39 × 10⁻³, 0.71 × 10⁻³], Fig. 1a). This upward trend was widespread, with 68% of grassland areas showing increasing NDVI_max trends since the early 1980s, including 45% experiencing significant increases (Fig. 1c). Prominent regions with rising trends included the Eurasian Steppes, eastern North American Great Plains, northern and central South American Pampas, and African grasslands (Fig. 1c). However, two abrupt shifts were detected in 1998 and 2009 using the BFAST method (Supplementary Fig. 1). Between these years, global grasslands experienced a reversal, with NDVI_max declining at a rate of −0.69 ± 0.40 × 10⁻³ yr⁻¹ (t (10) = −1.75, P = 0.11, 95% CI [−1.56 × 10⁻³, 0.19 × 10⁻³], Fig. 1a). During this period, 54% of grassland areas showed a downward NDVI_max trend, including 20% with significant declines, particularly in the eastern Eurasian Steppes (Fig. 1e). These findings demonstrate that 1998-2009 marked a period of interruption in the long-term increasing trend of grassland growth, indicating a stagnation in NDVI_max during this time.

**Fig. 1: Trend in NDVI_max and GPP_max across global grasslands.**

To validate the observed NDVI_max trends, we developed a globally continuous GPP dataset (1980–2020) by applying a machine learning method to upscale eddy covariance flux measurements from 75 grassland sites worldwide. Our analysis reveals a significant increasing trend in GPP_max from 1982 to 2020 at a rate of 3.63 ± 0.07 × 10⁻³ g C m⁻² d⁻¹ yr⁻¹ (t (37) = 4.96, P < 0.001, 95% CI [2.15 × 10⁻³, 5.12 × 10⁻³], Fig. 1b), with 60% of grassland areas exhibiting increasing GPP_max trends during this period (Fig. 1d). However, from 1998 to 2009, a notable decline in GPP_max was observed at a rate of −6.27 ± 0.31 × 10⁻³ g C m⁻² d⁻¹ yr⁻¹ (t (10) = −2.02, P < 0.1, 95% CI [−13.2 × 10⁻³, 0.66 × 10⁻³], Fig. 1b), with 56% regions showing decreasing trends (Fig. 1f). These eddy covariance-derived results provide alternative validation of the observed 1998–2009 interruption in increasing grassland growth peak.

Similarly, GPP_{max_EC-LUE} also followed the same pattern with NDVI_max, showing an increasing trend from 1982 to 2018 at a rate of 1.13 ± 0.12 × 10⁻² g C m⁻² d⁻¹ yr⁻¹ (t (35) = 9.67, P < 0.001, 95% CI [0.89 × 10⁻², 1.36 × 10⁻²]), but declining during the 1998-2009 period at a rate of −0.51 ± 0.29 × 10⁻² g C m⁻² d⁻¹ yr⁻¹ (t (10) = −1.77, P = 0.11, 95% CI [−1.14 × 10⁻², 0.13 × 10⁻²], Supplementary Fig. 2c). Moreover, EVI_max and SIF_max also demonstrated significant upward trends over the entire period (Supplementary Figs. 2d, e). However, the increasing rate for EVI_max and SIF_max during 2000-2009 declined to 14.5% and 38.7% of the respective rates observed across the overall period (Supplementary Figs. 2d, e). Furthermore, 77% of grassland areas exhibited an upward trend in GPP_{max_EC-LUE} from 1982 to 2018, with 53% experiencing a declining trend from 1998 to 2009 (Supplementary Fig. 3e, f). The declining trend area over 2000-2009 increased by 82% and 66% for EVI_max and SIF_max, respectively (Supplementary Figs. 3h, j). These findings further support our conclusion that the upward trajectory of grassland growth peaks was disrupted during 1998-2009, as evidenced by the decline in GPP_max and the marked slowdown in the increasing rates of EVI_max and SIF_max during this period. This reversal in multiple vegetation productivity indicators highlights a temporary stagnation in grassland peak growth, interrupting the long-term greening trend.

Regional contributions to trends in global grassland growth peak

At the regional level, 71% (20 out of 28) of IPCC climate regions exhibited an increasing NDVI_max trend over the entire 1982-2021 period (Supplementary Fig. 4). However, during the 1998–2009 period, 64% (18 out of 28) of regions showed decreasing NDVI_max trend, further confirming a widespread regional-scale interruption (Fig. 2). Similarly, 79% (22 out of 28) of regions showed an increasing trend of GPP_max based on eddy covariance method over 1982–2020, yet 54% (15 out of 28) exhibited a decreasing GPP_max trend during 1998–2009 (Fig. 2, Supplementary Fig. 4).

**Fig. 2: NDVI_max and GPP_max trend during 1998–2009 in different IPCC climate regions.**

To evaluate regional contributions to the global trend, we ranked regions based on area-weighted NDVI_max and GPP_max trends, considering the size of each area (Fig. 3). During the decline period of 1998–2009, the Tibetan Plateau (TIB) and East Asia (EAS) were the primary contributors to the global downturn, evidenced by both the NDVI_max and GPP_max trends (Fig. 3a, c). Notably, some regions, such as Central North America (CNA), Central America/Mexico (CAM), and West Asia (WAS) regions, maintained increasing NDVI_max and GPP_max trends during the interruption period (Fig. 3a, c). However, these increases were insufficient to counteract the overall global decline. A comparison of GPP_max and NDVI_max trends across IPCC regions revealed a strong correlation (t (25) = 3.75, R² = 0.56, P < 0.001, Supplementary Fig. 5), demonstrating their consistency. We identified 17 high-confidence regions where NDVI_max and GPP_max showed consistent trends (Supplementary Fig. 5). Notably, 11 regions, including TIB, EAS, and North Asia (NAS), exhibited significant declines in grassland growth peaks, supporting our conclusion of disrupted productivity. In contrast, regions like CNA, WAS, and CAM showed increasing trends, highlighting spatial heterogeneity in grassland responses.

**Fig. 3: Area weighted trend of NDVI_max and GPP_max.**

Decadal drought drives decline in global grassland growth peaks

From 1998 to 2009, global grasslands experienced a decadal drought characterized by a sharp decline in precipitation at a rate of −5.34 ± 1.4 mm yr⁻¹ (t (10) = −3.79, 95% CI [−8.49, −2.20], Fig. 4c). This decrease was four times faster than the rate from 1982 to 2021 (P < 0.05, Fig. 4c). During this period, over 68% of global grassland areas exhibited negative precipitation trends (Supplementary Fig. 6b), highlighting the widespread nature of the drought. Consistent with the precipitation decline, the standardized precipitation evapotranspiration index (SPEI) declined 3.4 times more rapidly during 1998-2009 (−4.04 × 10⁻² yr⁻¹; t (10) = −3.69, P < 0.001, 95% CI [−6.45 × 10⁻², −1.60 × 10⁻²]) compared to the entire period of 1982–2021 (−1.18 × 10⁻² yr⁻¹; t (10) = −5.79, P < 0.01, 95% CI [−1.60 × 10⁻², −0.77 × 10⁻²]) (Supplementary Fig. 7a). Similarly, extreme drought intensity decreased 5.3 times faster during 1998–2009 than 1982–2021 (−1.01 × 10⁻² yr⁻¹ vs. −0.19 × 10⁻² yr⁻¹, Supplementary Fig. 7b). In addition, the area affected by extreme drought expanded at a rate 2.4 times faster during 1998–2009 compared to the full study period (Supplementary Fig. 7c). These concurrent changes confirm the occurrence of a widespread decadal-scale drought during 1998–2009. Meanwhile, air temperature (TA), shortwave radiation (Rad), atmospheric CO₂ concentration, and nitrogen deposition rate (N_dep) showed significant linear increases throughout 1982–2021, including during the interruption period of 1998–2009 (Fig. 4a, e, g, i). Furthermore, cropland and rangeland fractions exhibited significant declining trends from 1998 to 2009 (Fig. 4k, m).

**Fig. 4: Trends of environmental variables and their partial correlation with NDVI_max.**

Partial correlation analysis revealed a strong positive relationship between detrended precipitation (PPT) and NDVI_max across 75% of global grasslands (Fig. 4d). Similarly, GPP_max exhibited positive partial correlations with PPT across 70% of global grasslands (Supplementary Fig. 9b). Among all environmental factors, PPT exhibited the highest mean partial correlation with GPP_max across global grasslands (Supplementary Fig. 9h). Therefore, the reduction in PPT during 1998–2009 was likely a major contributor to the observed declines in NDVI_max and GPP_max. To disentangle and quantify the contributions of various factors to NDVI_max and GPP_max changes, we further conducted factorial experiments using a machine learning model (see Methods). Our results indicate that in the absence of climate effects (S_CLI), NDVI_max and GPP_max continued to show an increasing trend during 1998–2009, highlighting climate change as the primary driver of NDVI_max and GPP_max decline (Fig. 5a, c). Among climate variables, precipitation exerted the strongest negative impact on NDVI_max and GPP_max, outweighing the positive effects of CO₂, nitrogen, cropland, and rangeland changes (Fig. 5b, d). Specifically, PPT changes led to declines in NDVI_max and GPP_max across 65% and 64% of global grasslands, respectively (Supplementary Figs. 10d, 11d). Rad also negatively affected both NDVI_max and GPP_max, consistent with its overall negative correlation with these indicators (Fig. 4f, 5b, d). The effects of temperature (TA) on global grassland NDVI_max and GPP_max are, on average, close to zero (Fig. 5b, d). Specifically, TA positively affected 51.0% and 49.7% of the global grassland area during 1982-2021 and 1998-2009, respectively (Supplementary Figs. 10b, 11b). In contrast, rising CO₂ concentrations and nitrogen deposition had a positive influence on NDVI_max and GPP_max during 1998–2009, partially offsetting the negative climate effects (Fig. 5b, d, Supplementary Figs. 10h, j and 11h, j). Additionally, declines in cropland and rangeland areas also had positive affect on NDVI_max and GPP_max (Fig. 5b, d, Supplementary Figs. 10l, n and Supplementary Figs. 11l, n). However, these positive effects were insufficient to offset the overall impact of climate variability, particularly the effects of decadal drought during 1998–2009.

**Fig. 5: NDVI_max and GPP_max simulations, and environmental contributions.**

We identified the dominant factor influencing NDVI_max as the one with the highest absolute partial correlation coefficient. In regions such as the Tibetan Plateau (TIB), East Asia (EAS), Northern Asia (NAS), i.e., key contributors to the decline in NDVI_max during 1998–2009, PPT was the primary factor, accounting for 37.4%, 41.8%, 27.0% of the respective areas (Fig. 6b). These regions also showed a decreasing trend in PPT (Supplementary Figs. 6b, 13). Moreover, these regions experienced highest extreme drought frequency during 1998-2009 (Supplementary Fig. 8c), with significantly higher extreme drought intensity in TIB and EAS (Supplementary Fig. 8d). Furthermore, during this decline period, we identified the dominant factor inducing the greatest GPP_max change, with PPT again emerging as the primary driver across global grasslands (Fig. 7a). In high-confidence regions with decreasing NDVI_max and GPP_max trend during 1998–2009, such as the TIB, ESA, NAS, South Asia (SAS), South Australia/New Zealand (SAU), and West Africa (WAF), the decline was likely driven primarily by decreasing precipitation (Fig. 7b, Supplementary Fig. 12). Conversely, in high-confidence regions where NDVI_max and GPP_max increased, such as Central America/Mexico (CAM), North-East Brazil (NEB), and Small islands regions: Caribbean (CAR) regions, the trend was likely primarily driven by rising precipitation (Fig. 7b, Supplementary Fig. 12). However, in Central North America (CNA) and West Asia (WAS), the increase in NDVI_max and GPP_max was likely influenced by rising CO₂ concentrations and nitrogen deposition. Additionally, land management practices, such as a reduction in cropland, also contributed to these observed changes (Fig. 7b, Supplementary Fig. 12). Moreover, we found that temperature played a significant role in high-latitude regions. In areas such as the Arctic (ARC) and Central Greenland (CGI), temperature was likely the primary controlling factor (Fig. 6b), with rising temperatures promoting increases in both GPP_max and NDVI_max (Fig. 7b, Supplementary Fig. 12). These results underscore the spatial heterogeneity in the dominant environmental controls; however, they do not change the overarching conclusion that decadal droughts have disrupted the peak growth of grasslands.

**Fig. 6: The dominant factors influencing variations in NDVI_max.**

**Fig. 7: The dominant factors of GPP_max and the effects of environmental variables during 1998-2009.**

Discussion

Our study identifies a decade-long stagnation in grassland growth peaks from 1998 to 2009, interrupting an otherwise upward trend observed from 1982 to 2021. This disruption, driven by a sharp reduction in precipitation, underscores the vulnerability of grasslands to climate variability, particularly in water-limited regions. Although the prevailing global greening narrative suggests a continuous increase in vegetation productivity^6,8,9, our findings reveal a critical yet overlooked shift in which global grasslands experienced a prolonged decline even as the overall trend continued to rise. While global greening trends have often masked such regional declines in vegetation growth^29,30,31, our findings highlight a distinct reversal affecting over 64% of IPCC climate regions, with significant impacts observed in the Tibetan Plateau and East Asia. Similar patterns have been noted in some regions, where short-term declines in vegetation growth are often preceded by periods of significant greening³². For example, regional studies across Eurasia have similarly reported a decrease in growing season NDVI in temperate and boreal ecosystems from the late 1990s to the early 2000s amidst a broader rising trend since the early 1980s^15,33. This drought-driven decrease in growth peaks aligns with reports of weakened ecosystem carbon sinks during the early 2000s, suggesting that grassland carbon storage capacity may be compromised under prolonged drought conditions^34,35.

Grasslands are highly sensitive to water availability and less resilient to drought^17,24,36,37, which implies heightened vulnerability to decadal droughts. Recent studies further highlight that, among various ecosystems, grasslands exhibit the highest drought sensitivity in terms of productivity over the past four decades^24,37. This heightened sensitivity is particularly evident in our findings, which reveal a decade-long stagnation in grassland growth peaks from 1998 to 2009, driven primarily by reduced precipitation. In contrast, forests are generally more resilient to water-deficient conditions and benefit more substantially from CO₂ fertilization^38,39. Global crop growth has also outpaced natural vegetation due to expanded agricultural areas and extended growing seasons^6,40. However, grasslands faced heightened risks from the amplified precipitation variability expected under global warming, making their stability and sustainability more precarious⁴¹.

The observed reversal in grassland growth peaks was closely tied to a decadal drought characterized by sharp declines in precipitation (Figs. 4, 5). Over the entire period from 1982 to 2021, NDVI_max and GPP_max trends benefited from rising CO₂, nitrogen deposition, contraction in cropland area, and reduction in grazing pressure (Fig. 5). Previous studies have similarly identified CO₂ fertilization and nitrogen deposition as key contributors to increasing global vegetation growth^6,8,9. As overgrazing and intensive agricultural management practices reduced vegetation production, the reduction in cropland and grazing pressure benefited the vegetation recovery^42,43. However, the intensification of decadal drought during 1998-2009 outweighed these positive effects, leading to a prolonged stagnation in grassland growth (Fig. 5). Experimental and field studies support this, with drought shown to reduce grassland productivity by 16% to 90%^18,44. Furthermore, recent global-scale experiments reveal that these impacts may be substantially underestimated, particularly in drier ecosystems with lower species diversity³⁶. This suggests water scarcity may drive more severe vegetation growth limitations than previously recognized, as evidenced by slower global vegetation dynamics during the 2000s^5,31. We also found that increasing solar radiation during drought periods negatively affected grassland growth peaks (Fig. 5). Enhanced net radiation, likely due to reduced cloud cover during drought, can exacerbate water stress and amplify heat extremes, ultimately reducing vegetation productivity⁴⁵. As future droughts intensify, understanding the mechanisms behind changes in vegetation growth peaks and their effects on grassland productivity becomes increasingly important.

Although the decadal drought reduced global grassland growth, regional heterogeneity emerged due to varying climate trends and ecosystem responses. The Tibetan Plateau (TIB) and East Asia (EAS) emerged as dominant contributors to the decline in NDVI_max and GPP_max during 1998–2009 (Figs. 2, 3). Regional studies confirm similar trends: vegetation growth in the TIB region generally increased from 1981 to 2020 but reversed during the 2000s as precipitation declined^46,47. Beyond TIB and EAS, regions like Northern Asia (NAS), South Asia (SAS), South Australia/New Zealand (SAU), and West Africa (WAF) also showed apparent declines, primarily driven by reductions in precipitation (Fig. 7b, Supplementary Fig. 12). Similarly, vegetation growth decline in South Australia has been reported in previous study⁴⁸. On the drylands in China, vegetation growth increased from 1982 to 2015, though NDVI declined between 1996 and 2005 due to decreasing precipitation¹⁵. The Arctic region also showed growth stagnation during 1998–2009, with extreme climate events slowing previously steady increases in Arctic vegetation productivity^49,50,51. In contrast, Central America/Mexico (CAM), Central North America (CNA), and West Asia (WAS) regions exhibited a sustained increase in vegetation growth peaks during 1998–2009, supported by localized environmental changes (Figs. 2, 7). Consistent with our findings, previous studies have reported persistent strong growth in Central North American grasslands from the 1980s through the 2010s^52,53. In CNA and WAS, positive trends were driven not only by rising CO₂ and nitrogen deposition but also by land management changes, especially cropland reduction (Fig. 7b, Supplementary Fig. 12). These land-use changes, particularly the conversion of croplands to grasslands, have been shown to enhance vegetation productivity and improve soil health⁵⁴. Taken together, despite individual regions varying, a consistent decadal reversal in global grassland growth peaks occurred from 1998 to 2009, interrupting the long-term upward trend observed from 1982 to 2021.

Our findings prove that decadal drought contributed to the interruption in the grassland growth peak. However, establishing definitive causal relationships remains challenging since our analysis is based on observational time-series data. Despite applying partial correlation analysis and factorial experiment through a machine learning model to account for interdependencies among environmental variables, the strong correlations among climate factors (e. g., air temperature, precipitation, and solar radiation) make it challenging to fully isolate their individual effects on grassland growth peak^55,56,57. Future experimental or process-based modeling studies will be valuable in elucidating the underlying mechanisms^18,58. While the GIMMS-4g NDVI dataset provides one of the most reliable long-term records of vegetation dynamics, it has limitations. The AVHRR satellite observations underlying this dataset may contain gaps and inconsistencies, particularly in the early years, due to sensor transitions and differences in satellite platforms⁵⁹. Although extensive calibration with MODIS NDVI has improved the temporal consistency of GIMMS NDVI²⁵, residual uncertainties remain, as inter-satellite discrepancies cannot be entirely eliminated⁶⁰. To account for these uncertainties, we cross-validated our results using multiple independent datasets, including GPP_max, EVI_max, and SIF_max. The general agreement among these datasets supports the robustness of our conclusions, though some regional discrepancies were observed. For instance, NDVI_max and GPP_max exhibited divergent trends in regions with higher uncertainty, suggesting that NDVI_max may not fully capture ecosystem productivity variations under certain conditions. These limitations should be considered when interpreting long-term trends in vegetation dynamics, particularly in regions with complex environmental influences.

In conclusion, our findings reveal a long-term rise in grassland growth peaks from 1982 to 2021, with a pronounced decade-long interruption from 1998 to 2009. During this period, 64% of IPCC climate regions, especially the Tibetan Plateau and East Asia, exhibited declines in growth peaks due to decadal drought with sharp declines in precipitation. Earth system models project an increasing frequency of decadal droughts alongside intensifying long-term severe droughts under global warming^61,62, such interruptions may increasingly offset gains in vegetation productivity. Given the critical role of grassland in regulating the interannual variability of the global carbon sink and supporting global food systems^1,3,4, the decadal droughts can temporarily pose significant risks to long-term positive trends in ecosystem function. Furthermore, the vegetation growth peak is one of the significant sources of uncertainty in terrestrial carbon cycle projections among the state-of-the-art Earth system models^63,64. Thus, a deeper understanding of vegetation growth peaks in response to decadal droughts is urgently needed for grassland management and climate change projections. Our results also highlight the need to incorporate decadal climate variability into ecosystem management, particularly as the impacts of extreme events increasingly shape ecosystem dynamics.

Methods

Dataset for vegetation growth peaks calculation

GIMMS-4g NDVI dataset

We utilized the biweekly GIMMS fourth-generation Normalized Difference Vegetation Index (GIMMS-NDVI 4 g; 1982–2021) datasets with a spatial resolution of 1/12° (~8 km) (https://doi.org/10.5281/zenodo.7649107). The GIMMS-NDVI4g dataset was consolidated from Advanced Very High-Resolution Radiometer (AVHRR) and Moderate-Resolution Imaging Spectroradiometer (MODIS) data. This version addresses limitations of previous GIMMS NDVI products, such as orbital drift and sensor degradation, providing a reliable long-term NDVI time series spanning four decades²⁵. The biweekly data were composited to a monthly temporal resolution by selecting the higher of the two composites within the same month. These monthly composites were then aggregated to a spatial resolution of 0.1° × 0.1° using bilinear interpolation to match the resolution of meteorological data.

EC-LUE GPP dataset

We used a global gross primary productivity (GPP) dataset derived from the eddy covariance light use efficiency model (EC-LUE model)²⁷. The model incorporates key environmental drivers, including temperature, radiation, and water availability, to simulate GPP at a global scale. The model was calibrated and validated using eddy covariance measurements from the FLUXNET2015 dataset. The dataset is available at an 8-day interval with a spatial resolution of 0.05° × 0.05° (https://doi.org/10.6084/m9.figshare.8942336.v3). The 8-day GPP data were composited to a monthly temporal resolution by selecting the highest value within each month. The GPP data were then aggregated to a spatial resolution of 0.1° × 0.1° using bilinear interpolation to match the resolution of meteorological data.

EVI dataset

We used the Enhanced Vegetation Index (EVI) data from the Terra MODIS Vegetation Indices 16-Day (MOD13C1) Version 6.1 product. This dataset optimizes the vegetation signal by reducing atmospheric cloud and aerosol contamination effects, providing a complementary proxy for interpreting peak vegetation growth⁶⁵. The 16-day MODIS EVI data, with a spatial resolution of 0.05° (2001–2021), were obtained from the NASA Land Processes Distributed Active Archive Center (https://lpdaac.usgs.gov). The biweekly data were composited to a monthly temporal resolution by selecting the higher of the two composites within the same month. The EVI data were then resampled to a spatial resolution of 0.1° × 0.1° using bilinear interpolation.

SIF dataset

We used the global spatially contiguous solar-induced chlorophyll fluorescence (CSIF) dataset, which has a 4-day temporal resolution from 2001 to 2018 and a spatial resolution of 0.5° × 0.5° (https://doi.org/10.6084/m9.figshare.6387494). The CSIF dataset was generated by training a neural network (NN) with surface reflectance data from MODIS and SIF measurements from the Orbiting Carbon Observatory-2 (OCO-2)²⁸. This approach addresses limitations of traditional SIF datasets, such as low spatial and temporal resolution, high uncertainty in individual retrievals, and inconsistencies in measurement footprints. The CSIF data were then aggregated to a spatial resolution of 0.1° × 0.1° using bilinear interpolation to match the resolution of meteorological data.

Datasets of environmental variables

Climate datasets

Monthly climatology data of solar radiation, precipitation, and air temperature (1982–2021) with a spatial resolution of 0.1° × 0.1° were obtained from the fifth-generation European Center for Medium-Range Weather Forecasts (ECMWF) atmospheric reanalysis (https://www.ecmwf.int/). Annual air temperature was calculated as the mean of monthly data, while annual precipitation and solar radiation were summed from monthly data.

Nitrogen deposition rate and CO₂ concentration dataset

Annual global nitrogen deposition rate, with a spatial resolution of 0.5° × 0.5°, was obtained from the TRENDY model environmental driver datasets (https://aims2.llnl.gov/search/input4mip). The nitrogen dataset was resampled to 0.1° × 0.1° using the bilinear interpolation. Global mean CO₂ concentration data was sourced from National Oceanic and Atmospheric Administration (NOAA) Global Monitoring Laboratory measurements (https://aims2.llnl.gov/search/input4mips/). The original 0.5° × 0.5° nitrogen data were aggregated to a spatial resolution of 0.1° × 0.1° using bilinear interpolation.

Land use change datasets

Annual global cropland fraction and rangeland fraction data were obtained from the Land-Use Harmonization (LUH2) dataset, with a spatial resolution of 0.25° × 0.25° (https://luh.umd.edu/). This dataset, developed as part of the World Climate Research Program Coupled Model Intercomparison Project (CMIP6), provides a harmonized set of land-use scenarios for Earth System Models (ESMs)⁶⁶. The data were resampled to 0.1° × 0.1° using bilinear interpolation.

Land cover and IPCC regional classification datasets

We used the MODIS Land Cover Climate Modeling Grid (CMG) (MCD12C1) Version 6.1 data product to define grassland areas (https://lpdaac.usgs.gov/products/mcd12c1v061/). This dataset provides global land cover maps at a spatial resolution of 0.05° (~5.6 km), based on the International Geosphere-Biosphere Program (IGBP) classification scheme. We adopted the regional classification from the IPCC Atlas of Global and Regional Climate Projections (https://www.ipcc-data.org/guidelines/pages/ar5_regions.html). These regions were originally defined to represent consistent climatic regimes and physiographic settings while maintaining an appropriate spatial scale⁶⁷. The base maps of continental boundaries presented in the figures were obtained from Natural Earth (www.naturalearthdata.com) using the R package “rnaturalearth”.

Analysis

Generating of global grassland GPP using eddy covariance observation

We collected flux data from 75 grassland sites worldwide and upscaled them to a global scale at a monthly resolution using the XGBoost method (Supplementary Table R1). Among them, 68 sites were obtained from FLUXNET, AmeriFLUX (https://ameriflux.lbl.gov/), the Integrated Carbon Observation System (ICOS, https://www.icos-cp.eu/), and ChinaFLUX (https://chinaflux.org/). In addition, we included our own observations of 7 sites in Inner Mongolia grasslands (Supplementary Table 1). We selected GPP data derived from the nighttime flux partitioning method^68,69. Finally, the dataset consists of 5296 site-month observations, providing global grassland GPP estimates from 1982 to 2020 at a spatial resolution of 0.1°. The XGBoost model was trained to estimate carbon fluxes based on key meteorological and satellite-derived variables, including air temperature (TA), precipitation (PPT), vapor pressure deficit (VPD), solar radiation (Rad), and leaf area index (LAI). The monthly Rad, TA, PPT, and VPD were derived from the ERA5 dataset. The LAI dataset (1982–2020) was obtained from the GIMMS-LAI 4 g dataset (https://doi.org/10.5281/zenodo.7649107). The original biweekly LAI data at 1/12° resolution were averaged to generate monthly values and resampled to 0.1° resolution using bilinear interpolation for consistency with other datasets. The specific explanatory variables for the estimation of carbon fluxes were classified into three types: spatial; spatial and seasonal; spatial, seasonal, and interannual variables (Supplementary Table 2). To evaluate the performance of the XGBoost model, we conducted cross-validation by randomly dividing the flux data into two groups: 90% (4766 site-months) was used as the training dataset, while the remaining 10% (527 site-months) served as the validation dataset. We used Grid Search to perform hyperparameter tuning for our machine learning model. The specific hyperparameters explored during the grid search are provided in Supplementary Table 3. Then, the trained XGBoost model was applied to the validation dataset to test the model performance in simulating carbon fluxes. The XGBoost model demonstrated strong predictive performance, achieving an R² of 0.98 for the training data and 0.92 for the validation data (Supplementary Fig. 14). Finally, we obtained the time-varying estimation of carbon fluxes in each pixel by applying the well-trained XGBoost on gridded input datasets with 0.1° × 0.1° spatial resolution and monthly time step from the period of 1982–2020. For hyperparameter optimization, we employed the “GridSearchCV” function from the “scikit-learn package” in Python 3.10.9. The XGBoost model was implemented using the “xgboost” package.

Grassland peak growth calculation

Annual gridded NDVI_max, GPP_{max_Eddy}, GPP_{max_EC-LUE}, EVI_max, and SIF_max datasets were compiled by calculating the maximum value from annual composites for each grid cell. Data processing was performed using the “ncdf4”, “terra”, and “raster” packages in R 4.5.0.

Detection of downward period

To detect and characterize the abrupt changes (i.e., breakpoints) in the global grassland NDVI_max time series, we adopted the Breaks For Additive Season and Trend (BFAST) method²⁶. This method detects structural changes in the trend based on the moving sums (MOSUMs) test. The analysis was implemented using the “bfast” package in R 4.5.0.

Linear trend calculation

Linear trends in NDVI_max, GPP_max, GPP_{max_EC-LUE}, EVI_max, and SIF_max were calculated for the entire study period (1982–2021/2018 or 2000–2018/2021) and specific intervals (1998–2009 or 2001–2009). Spatial patterns of temporal trends were derived by fitting least squares linear regression models at each grid cell, with the resulting spatial patterns representing these individual grid-cell-level trends.

Area-weighted mean value calculated for global and regional scales

The area-weighted value for a given variable is computed as:

$${{\mbox{X}}}_{{weighted}}=\frac{\sum ({{\mbox{X}}}_{{\mbox{i}}}\times {{\mbox{A}}}_{{\mbox{i}}})}{\sum {{\mbox{A}}}_{{\mbox{i}}}}$$

(1)

where X_i is the variable value in the i-th grid cell, A_i is the area of the i-th grid cell. For the global average calculation, the summation is performed across all grid cells within the global domain. For regional trend analysis, we normalized the regional contributions to the global trend by the total global area $\sum {A}_{i}$.

Partial correlation calculation

We used the partial correlation method to analyze the correlation between detrended environmental variables and NDVI_max at each pixel. Before the analysis, the environmental variables and NDVI_max were detrended to remove long-term trends. Detrending was performed by fitting a linear regression model to the time series of each variable at each pixel and subtracting the fitted trend from the original data. The data from each grid cell and year combination is considered an independent observation. The partial correlation analysis was conducted to isolate the influence of individual environmental factors on NDVI_max while controlling for the effects of other correlated variables⁵⁵. The analysis was implemented using the “ppcor” package in R 4.5.0.

Factorial experiment through machine learning model

We developed Random Forest (RF) models to simulate NDVI_max and GPP_max at each pixel by precipitation (PPT), air temperature (TA), solar radiation (Rad), CO₂ concentration (CO₂), nitrogen deposition rate (N), cropland fraction (Cropland), and rangeland fraction (Rangeland)⁵⁷. The RF model was driven by all time-varying variables (S_ALL) and supplemented with five factorial simulations (S_CLI, S_CO2, S_N, S_Cropland, and S_Rangeland), where each simulation held a single factor (climate, CO₂, nitrogen deposition rate, cropland fraction, or rangeland fraction) constant at its initial level (first year of data) while allowing the other variables to vary with time. The differences between the all-variable-varying simulation (S_ALL) and the other factorial simulations (S_CLI, S_CO2, S_N, S_Cropland, and S_Rangeland) were used to estimate the sensitivity of NDVI_max and GPP_max to each variable (Eq. (2)~(3)).

$${\triangle {NDVI}({GPP})}_{\max \left({{S}}_{{ALL}}-{{S}}_{{CLI}}\right)}= {{{\beta }}}_{{PPT}}\times {\triangle {PPT}}_{({{S}}_{{ALL}}-{{S}}_{{CLI}})}+{{{\beta }}}_{{TA}}\times {\triangle {TA}}_{({{S}}_{{ALL}}-{{S}}_{{CLI}})} \\ +{{{\beta }}}_{{Rad}}\times {\triangle {Rad}}_{({{S}}_{{ALL}}-{{S}}_{{CLI}})}+{{\varepsilon }}$$

(2)

$${\triangle {NDVI}({GPP})}_{\max \left({{S}}_{{ALL}}-{{S}}_{{{i}}}\right)}={{{\beta }}}_{{i}}\times {\triangle {X}}_{({{S}}_{{ALL}}-{{S}}_{{i}})}+{{\varepsilon }}$$

(3)

where ${\triangle {NDVI}({GPP})}_{\max }$, $\triangle {PPT}$, $\triangle {TA}$, and $\triangle {Rad}$ represent the differences of PPT, TA, and Rad between two different simulation types from 1982–2021/2020, and ε is the residual error. ${{\beta }}_{{PPT}}$, ${{\beta }}_{{TA}}$, ${{\beta }}_{{Rad}}$, represents the sensitivity of NDVI_max (GPP_max) to different variables. In Eq. (3), S_i represents the reference condition where a specific factor is held constant, and i corresponds to CO₂ concentration, nitrogen deposition rate, cropland fraction, and rangeland fraction. X denotes the specific environmental factor associated with i. We quantified the contributions of these variables to NDVI_max and GPP_max during 1982–2021 and 1998–2009 by multiplying the magnitude of their change rate and sensitivity of NDVI_max (GPP_max) (β).

Identification of extreme drought

The global monthly SPEI data was integrated at the 12-month time scale (hereafter SPEI-12) (available at https://spei.csic.es/database.html) to quantify extreme drought events. Based on the annual SPEI-12 data, we defined the thresholds for annual extreme drought events as those dry events (negative SPEI values) occurring less frequently than once per decade (10% of observations) over the past century (1901–2021) at each grid cell⁷⁰. The grid-specific SPEI-12 thresholds were used to screen for extreme drought events, and therefore, 10% of the SPEI historic climate years (120 years) were identified as extreme drought events. The global mean annual drought intensity was calculated as the mean of the SPEI-12 values at all grid cells experiencing extreme drought conditions in that year. The annual drought area was calculated by summing up the area of all grid cells identified as experiencing extreme drought. The drought frequency at each grid cell is defined as the number of years from 1982 to 2021, during which the grid cell experienced an extreme drought event.

Reporting summary

Further information on research design is available in the Nature Portfolio Reporting Summary linked to this article.

Data availability

The GPP_max dataset generated in this study is available on Figshare at https://doi.org/10.6084/m9.figshare.29924594. The GIMMS-4g NDVI that supports the findings of this study is available at https://doi.org/10.5281/zenodo.7649107. The EC-LUE GPP dataset is available at https://doi.org/10.6084/m9.figshare.8942336.v3. MODIS EVI dataset is available at https://lpdaac.usgs.gov/products/mod13c1v061/. SIF dataset is available at https://figshare.com/articles/dataset/CSIF/6387494/2. ERA5 climate dataset is available at https://www.ecmwf.int/. The GIMMS-4g LAI dataset is available at https://doi.org/10.5281/zenodo.7649107. Nitrogen deposition dataset is available at https://aims2.llnl.gov/search/input4mips/. MODIS Land cover dataset is available at https://lpdaac.usgs.gov/products/mcd12c1v061/. IPCC regional classification dataset is available at https://www.ipcc-data.org/guidelines/pages/ar5_regions.html. SPEI dataset is available at https://spei.csic.es/database.html. Source data are provided with this paper.

Code availability

Codes are available on GitHub at https://github.com/CuihaiYou/gragrassland-growth-peak-code and on Code Ocean at https://doi.org/10.24433/CO.4190630.v1.

References

Liu, L. et al. The grassland carbon cycle: mechanisms, responses to global changes, and potential contribution to carbon neutrality. Fund. Res. 3, 209–218 (2023).
CAS Google Scholar
Stevens, N., Bond, W., Feurdean, A. & Lehmann, C. E. R. Grassy ecosystems in the Anthropocene. Annu. Rev. Environ. Resour. 47, 261–289 (2022).
Article Google Scholar
Ahlstrom, A. et al. The dominant role of semi-arid ecosystems in the trend and variability of the land CO₂ sink. Science 348, 895–899 (2015).
Article ADS PubMed Google Scholar
Poulter, B. et al. Contribution of semi-arid ecosystems to interannual variability of the global carbon cycle. Nature 509, 600–603 (2014).
Article ADS PubMed CAS Google Scholar
Zhang, A., Jia, G. & Ustin, S. L. Water availability surpasses warmth in controlling global vegetation trends in the recent decade: revealed by satellite time series. Environ. Res. Lett. 16, 074028 (2021).
Article ADS CAS Google Scholar
Huang, K. et al. Enhanced peak growth of global vegetation and its key mechanisms. Nat. Ecol. Evol. 2, 1897–1905 (2018).
Article PubMed Google Scholar
Xia, J. et al. Joint control of terrestrial gross primary productivity by plant phenology and physiology. Proc. Natl Acad. Sci. 112, 2788–2793 (2015).
Article ADS PubMed PubMed Central CAS Google Scholar
Piao, S. et al. Characteristics, drivers and feedbacks of global greening. Nat. Rev. Earth Environ. 1, 14–27 (2020).
Article ADS Google Scholar
Zhu, Z. et al. Greening of the Earth and its drivers. Nat. Clim. Change 6, 791–795 (2016).
Article ADS CAS Google Scholar
Reeves, M. C., Hanberry, B. B., Wilmer, H., Kaplan, N. E. & Lauenroth, W. K. An assessment of production trends on the Great Plains from 1984 to 2017. Rangel. Ecol. Manag. 78, 165–179 (2021).
Article Google Scholar
Miao, L., Sun, Z., Ren, Y., Schierhorn, F. & Mueller, D. Grassland greening on the Mongolian Plateau despite higher grazing intensity. Land Degrad. Dev. 32, 792–802 (2021).
Article Google Scholar
Wu, S., Gao, X., Lei, J., Zhou, N. & Wang, Y. Spatial and temporal changes in the Normalized Difference Vegetation Index and their driving factors in the desert/grassland biome transition zone of the Sahel region of Africa. Remote Sens. 12, 4119 (2020).
Article ADS Google Scholar
Miao, L. et al. Modeling with hysteresis better captures grassland growth in Asian drylands. Remote Sens. 16, 1838 (2024).
Article ADS Google Scholar
Scotta, F. C. & da Fonseca, E. L. Multiscale trend analysis for Pampa grasslands using ground data and vegetation sensor imagery. Sensors 15, 17666–17692 (2015).
Article ADS PubMed PubMed Central Google Scholar
Guo, X. et al. The effects of precipitation event characteristics and afforestation on the greening in arid grasslands, China. Remote Sens. 12, 4119 (2023).
ADS Google Scholar
Yu, Q. et al. Contrasting drought sensitivity of Eurasian and North American grasslands. Nature 63, 114–118 (2025).
Article ADS Google Scholar
Song, J. et al. A meta-analysis of 1,119 manipulative experiments on terrestrial carbon-cycling responses to global change. Nat. Ecol. Evol. 3, 1309–1320 (2019).
Article PubMed Google Scholar
Knapp, A. K. et al. Field experiments have enhanced our understanding of drought impacts on terrestrial ecosystems—But where do we go from here?. Funct. Ecol. 38, 76–97 (2024).
Article CAS Google Scholar
Zhang, W., Zhou, T. & Wu, P. Anthropogenic amplification of precipitation variability over the past century. Science 385, 427–432 (2024).
Article ADS PubMed CAS Google Scholar
Yuan, X. et al. A global transition to flash droughts under climate change. Science 380, 187–191 (2023).
Article ADS PubMed CAS Google Scholar
Schwalm, C. R. et al. Reduction in carbon uptake during the turn of the century drought in western North America. Nat. Geosci. 5, 551–556 (2012).
Article ADS CAS Google Scholar
Dong, C. et al. Vegetation responses to 2012–2016 drought in Northern and Southern California. Geophys. Res. Lett. 46, 3810–3821 (2019).
Article ADS Google Scholar
Yuan, W. et al. Multiyear precipitation reduction strongly decreases carbon uptake over northern China. J. Geophys. Res.-Biogeosci. 119, 881–896 (2014).
Article CAS Google Scholar
Chen, L. et al. Global increase in the occurrence and impact of multiyear droughts. Science 387, 278–284 (2025).
Article ADS PubMed CAS Google Scholar
Li, M. et al. Spatiotemporally consistent global dataset of the GIMMS Normalized Difference Vegetation Index (PKU GIMMS NDVI) from 1982 to 2022. Earth Syst. Sci. Data 15, 4181–4203 (2023).
Article ADS Google Scholar
Verbesselt, J., Hyndman, R., Newnham, G. & Culvenor, D. Detecting trend and seasonal changes in satellite image time series. Remote Sens. Environ. 114, 106–115 (2010).
Article ADS Google Scholar
Zheng, Y. et al. Improved estimate of global gross primary production for reproducing its long-term variation, 1982–2017. Earth Syst. Sci. Data 12, 2725–2746 (2020).
Article ADS Google Scholar
Zhang, Y., Joiner, J., Alemohammad, S. H., Zhou, S. & Gentine, P. A global spatially contiguous solar-induced fluorescence (CSIF) dataset using neural networks. Biogeosciences 15, 5779–5800 (2018).
Article ADS CAS Google Scholar
Madani, N. et al. Recent amplified global gross primary productivity due to temperature increase is offset by reduced productivity due to water constraints. AGU Adv. 1, e2020AV000180 (2020).
Article ADS Google Scholar
Pan, N. et al. Increasing global vegetation browning hidden in overall vegetation greening: Insights from time-varying trends. Remote Sens. Environ. 214, 59–72 (2018).
Article ADS Google Scholar
Winkler, A. J. et al. Slowdown of the greening trend in natural vegetation with further rise in atmospheric CO₂. Biogeosciences 18, 4985–5010 (2021).
Article ADS CAS Google Scholar
Jong, R. D., Verbesselt, J., Schaepman, M. E. & Bruin, S. D. Trend changes in global greening and browning: contribution of short-term trends to longer-term change. Glob. Change Biol. 18, 642–655 (2012).
Article ADS Google Scholar
Piao, S. et al. Changes in satellite-derived vegetation growth trend in temperate and boreal Eurasia from 1982 to 2006. Glob. Change Biol. 17, 3228–3239 (2011).
Article ADS Google Scholar
You, C. et al. Inner Mongolia grasslands act as a weak regional carbon sink: a new estimation based on upscaling eddy covariance observations. Agric. For. Meteorol. 342, 109719 (2023).
Article Google Scholar
Yang, Y. et al. Changes in topsoil carbon stock in the Tibetan grasslands between the 1980s and 2004. Glob. Change Biol. 15, 2723–2729 (2009).
Article ADS Google Scholar
Smith, M. D. et al. Extreme drought impacts have been underestimated in grasslands and shrublands globally. Proc. Natl Acad. Sci. 121, e2309881120 (2024).
Article PubMed PubMed Central Google Scholar
Tang, J. et al. Shifted trend in drought sensitivity of vegetation productivity from 1982 to 2020. Agric. For. Meteorol. 362, 110388 (2025).
Article Google Scholar
Huang, K. et al. High ecosystem stability of evergreen broadleaf forests under severe droughts. Glob. Change Biol. 25, 3494–3503 (2019).
Article ADS Google Scholar
Pan, Y. et al. Contrasting responses of woody and grassland ecosystems to increased CO₂ as water supply varies. Nat. Ecol. Evol. 6, 315–323 (2022).
Article PubMed Google Scholar
Gao, X., Liang, S. B. & He, B. Detected global agricultural greening from satellite data. Agric. For. Meteorol. 276, 107652 (2019).
Article Google Scholar
Knapp, A. K., Ciais, P. M. & Smith, D. Reconciling inconsistencies in precipitation–productivity relationships: implications for climate change. N. Phytol. 214, 41–47 (2017).
Article Google Scholar
Buisson, E. et al. Ancient grasslands guide ambitious goals in grassland restoration. Science 377, 594–598 (2022).
Article ADS PubMed CAS Google Scholar
Yan, L., Zhou, G. & Zhang, F. Effects of different grazing intensities on grassland production in China: a meta-analysis. PLoS ONE 8, e81466 (2013).
Article ADS PubMed PubMed Central Google Scholar
Kröel-Dulay, G. et al. Field experiments underestimate aboveground biomass response to drought. Nat. Ecol. Evol. 6, 540–545 (2022).
Article PubMed PubMed Central Google Scholar
Zhou, S. et al. The role of climate and vegetation in regulating drought-heat extremes. Nat. Commun. 13, 3135 (2022).
ADS Google Scholar
Wang, Y. et al. Grassland changes and adaptive management on the Qinghai-Tibetan Plateau. Nat. Rev. Earth Environ. 3, 668–683 (2022).
Article ADS Google Scholar
Li, P., Hu, Z. & Liu, Y. Shift in the trend of browning in the Southwestern Tibetan Plateau in the past two decades. Agric. For. Meteorol. 287, 107950 (2020).
Article Google Scholar
Cook, B. I. & Pau, S. A global assessment of long-term greening and browning trends in pasture lands using the GIMMS LAI3g dataset. Remote Sens. 5, 2492–2512 (2013).
Article ADS Google Scholar
Phoenix, G. K. & Bjerke, J. W. Arctic browning: extreme events and trends reversing arctic greening. Glob. Change Biol. 22, 2960–2962 (2016).
Article ADS Google Scholar
Fu, D., Su, F., Wang, J. & Sui, Y. Patterns of Arctic tundra greenness based on spatially downscaled Solar-Induced Fluorescence. Remote Sens. 11, 1460 (2019).
Article ADS Google Scholar
Berner, L. T. et al. Summer warming explains widespread but not uniform greening in the Arctic tundra biome. Nat. Commun. 11, 4621 (2020).
Article ADS PubMed PubMed Central CAS Google Scholar
Liu, Y., Wu, C., Wang, X., Jassal, R. S. & Gonsamo, A. Impacts of global change on peak vegetation growth and its timing in terrestrial ecosystems of the continental US. Glob. Planet. Change 207, 103657 (2021).
Article Google Scholar
Nash, M. S., Wickham, J., Christensen, J. & Wade, T. Changes in landscape greenness and climatic factors over 25 years (1989-2013) in the USA. Remote Sens. 9, 295 (2017).
Article ADS Google Scholar
Fargione, J. E. et al. Natural climate solutions for the United States. Sci. Adv. 4, eaat1869 (2018).
Article ADS PubMed PubMed Central Google Scholar
Baba, K., Shibata, R. & Sibuya, M. Partial correlation and condition correlation as measures of conditional independence. Aust. N.Z. J. Stat. 46, 657–664 (2004).
Article MathSciNet Google Scholar
Runge, J. et al. Inferring causation from time series in Earth system sciences. Nat. Commun. 10, 2553 (2019).
Article ADS PubMed PubMed Central Google Scholar
Song, Y., Jiao, W., Wang, J. & Wang, L. Increased global vegetation productivity despite rising atmospheric dryness over the last two decades. Earth’s Future 10, e2021EF002634 (2022).
Article ADS Google Scholar
Tilman, D., Ecological experimentation: strengths and conceptual problems. In: G. E. Likens (Editor), Long-Term studies in ecology: approaches and alternatives. 136-157 (Springer New York, New York, NY, 1989).
Jeong, S. et al. Persistent global greening over the last four decades using novel long-term vegetation index data with enhanced temporal consistency. Remote Sens. Environ. 311, 114282 (2024).
Article Google Scholar
Tian, F. et al. Evaluating temporal consistency of long-term global NDVI datasets for trend analysis. Remote Sens. Environ. 163, 326–340 (2015).
Article ADS Google Scholar
Behzadi, F. et al. Projections of meteorological drought severity-duration variations based on CMIP6. Sci. Rep. 14, 5027 (2024).
Article ADS PubMed PubMed Central CAS Google Scholar
Li, H. et al. Drylands face the potential threat of robust drought in the CMIP6 SSPs scenarios. Environ. Res. Lett. 16, 114004 (2021).
Article ADS Google Scholar
Huang, K. & Xia, J. Global synchronous increase in light-saturated and peak vegetation productivity. Fundamental Research, https://doi.org/10.1016/j.fmre.2024.09.001 (2024).
Xia, J. et al. Terrestrial ecosystem model performance in simulating productivity and its vulnerability to climate change in the northern permafrost region. J. Geophys. Res.: Biogeosci. 122, 430–446 (2017).
Article CAS Google Scholar
Huete, A. et al. Overview of the radiometric and biophysical performance of the MODIS vegetation indices. Remote Sens. Environ. 83, 195–213 (2002).
Article ADS Google Scholar
Hurtt, G. C. et al. Harmonization of global land use change and management for the period 850–2100 (LUH2) for CMIP6. Geosci. Model Dev. 13, 5425–5464 (2020).
Article ADS CAS Google Scholar
Giorgi, F. & Francisco, R. Uncertainties in regional climate change prediction: a regional analysis of ensemble simulations with the HADCM2 coupled AOGCM. Clim. Dyn. 16, 169–182 (2000).
Article Google Scholar
Pastorello, G. et al. The FLUXNET2015 dataset and the ONEFlux processing pipeline for eddy covariance data. Sci. Data 7, 225 (2020).
Article PubMed PubMed Central Google Scholar
Reichstein, M. et al. On the separation of net ecosystem exchange into assimilation and ecosystem respiration: review and improved algorithm. Glob. Change Biol. 11, 1424–1439 (2005).
Article ADS Google Scholar
Huang, K. & Xia, J. High ecosystem stability of evergreen broadleaf forests under severe droughts. Glob. Change Biol. 10, 3494–3503 (2019).
Article ADS Google Scholar

Download references

Acknowledgments

We are grateful for receiving NDVI data from the GIMMS team, MODIS EVI and land cover products from the United States Geological Survey (USGS), and CO₂ data from the National Oceanic and Atmospheric Administration (NOAA). This study utilized climate data from the ERA-Interim reanalysis dataset provided by the European Center for Medium-Range Weather Forecasts (ECMWF). We also thank the TRENDY team for providing nitrogen deposition data. We are additionally grateful to Yao Zhang et al. for the SIF dataset and to Yi Zheng et al. for the EC-LUE GPP dataset. We gratefully acknowledge FLUXNET, AmeriFlux, ICOS, and ChinaFLUX for their efforts in collecting, processing, and sharing open-access eddy covariance data. We also acknowledge USCCC and ChinaFLUX for their methodological support in eddy covariance data processing. This work is financially supported by the National Natural Science Foundation of China (32325033), the National Key R&D Program of China (2022YFF0802104; 2023YFF1305301), and the National Natural Science Foundation of China (32401389; U22A20559). J.X. is supported by the Shanghai Pilot Program for Basic Research (TQ20220102).

Author information

These authors contributed equally: Cuihai You, Shiping Chen.

Authors and Affiliations

Research Center for Global Change and Ecological Forecasting, School of Ecological and Environmental Sciences, Institute of Eco-Chongming, East China Normal University, Shanghai, 200241, China
Cuihai You, Zhiqin Tu, Chenyu Bian, Erqian Cui, Kun Huang, Fangxiu Wan, Jiaye Ping & Jianyang Xia
Institute of Botany, Chinese Academy of Sciences, Beijing, 100093, China
Shiping Chen
School of Integrative Plant Science, College of Agriculture and Life Sciences, Cornell University, Ithaca, NY, 14853, USA
Ning Wei

Authors

Cuihai You
View author publications
Search author on:PubMed Google Scholar
Shiping Chen
View author publications
Search author on:PubMed Google Scholar
Zhiqin Tu
View author publications
Search author on:PubMed Google Scholar
Chenyu Bian
View author publications
Search author on:PubMed Google Scholar
Erqian Cui
View author publications
Search author on:PubMed Google Scholar
Kun Huang
View author publications
Search author on:PubMed Google Scholar
Fangxiu Wan
View author publications
Search author on:PubMed Google Scholar
Jiaye Ping
View author publications
Search author on:PubMed Google Scholar
Ning Wei
View author publications
Search author on:PubMed Google Scholar
Jianyang Xia
View author publications
Search author on:PubMed Google Scholar

Contributions

J.X. and C.Y. designed the study. C.Y. performed the data analysis and wrote the original manuscript. S.C. provided eddy covariance data from Inner Mongolia for machine learning and contributed to the writing. Z.T. and C.B. processed remote sensing data. E.C., K.H., F.W., N.W., and J.P. contributed to manuscript review and editing. J.X. conceived and supervised the research, secured funding, and finalized the manuscript. All authors contributed to the manuscript revisions.

Corresponding author

Correspondence to Jianyang Xia.

Ethics declarations

Competing interests

The authors declare no competing interests.

Peer review

Peer review information

Nature Communications thanks Midhun Mohan, who co-reviewed with Abhilash Dutta Roy and the other, anonymous, reviewer(s) for their contribution to the peer review of this work. A peer review file is available.

Additional information

Publisher’s note Springer Nature remains neutral with regard to jurisdictional claims in published maps and institutional affiliations.

Supplementary information

Supplementary Information

Peer Review file

Reporting Summary

Source data

Source Data

Rights and permissions

Open Access This article is licensed under a Creative Commons Attribution-NonCommercial-NoDerivatives 4.0 International License, which permits any non-commercial use, sharing, distribution and reproduction in any medium or format, as long as you give appropriate credit to the original author(s) and the source, provide a link to the Creative Commons licence, and indicate if you modified the licensed material. You do not have permission under this licence to share adapted material derived from this article or parts of it. The images or other third party material in this article are included in the article’s Creative Commons licence, unless indicated otherwise in a credit line to the material. If material is not included in the article’s Creative Commons licence and your intended use is not permitted by statutory regulation or exceeds the permitted use, you will need to obtain permission directly from the copyright holder. To view a copy of this licence, visit http://creativecommons.org/licenses/by-nc-nd/4.0/.

Reprints and permissions

About this article

Cite this article

You, C., Chen, S., Tu, Z. et al. Decadal trends in global grassland growth peaks and their drivers since the 1980s. Nat Commun 16, 9501 (2025). https://doi.org/10.1038/s41467-025-64565-x

Download citation

Received: 09 December 2024
Accepted: 19 September 2025
Published: 28 October 2025
Version of record: 28 October 2025
DOI: https://doi.org/10.1038/s41467-025-64565-x