Diminishing aerosol-related compensation of vegetation productivity across China: asymmetric responses of AGC, LAI, NPP, and GPP from 2002 to 2020

1 Introduction

Atmospheric aerosols refer to stable mixtures of tiny solid or liquid particles suspended in the atmosphere with aerodynamic diameters ranging from 0.001 to 100 μm, which have a significant environmental effect (Liu and He, 2007). They are complex in structure and have diverse sources. Aerosol optical depth (AOD) is defined as the integral of the extinction coefficient of the entire aerosol layer in the vertical direction. It is a pivotal parameter for characterizing aerosol properties (Zhou H et al., 2024) and has been extensively applied in domains such as environmental monitoring, atmospheric radiation transfer, remote sensing applications, and climate change (Wang et al., 2019). Atmospheric aerosols have a wide range of sources, including anthropogenic (industrial emissions, vehicle exhaust, biomass burning, etc.) and natural (volcanic ash, dust, sea spray, biogenic aerosols, etc.) sources. Primary aerosols generated by natural emissions or direct human activities undergo chemical reactions in the atmosphere and form secondary aerosols, which together affect various aspects of human life and the ecological environment (Li et al., 2025). Among them, atmospheric aerosols affect the amount of solar radiation reaching the ground and alter the proportion of scattered radiation, thereby promoting or inhibiting photosynthesis in vegetation (Zhou et al., 2021) and changing ecosystem productivity. Therefore, a comprehensive investigation into the statistical relationships and asymmetric responses of vegetation photosynthesis and ecosystem productivity to atmospheric aerosols is important for improving our understanding of aerosol–vegetation interactions and their ecological implications.

Two methodological frameworks prevail in aerosol-vegetation research, including direct monitoring (ground and satellite observations) and indirect modeling simulations. Station-based observation data is highly reliable and accurate, but there are issues such as a limited number of stations and high costs, making it difficult to support long-term, large-scale data observation tasks. Conversely, satellite remote sensing offers the capability to promptly, comprehensively, and consistently acquire data concerning aerosols and vegetation productivity. Concurrently, advancements in scientific research have led to continual improvements in the spatial and temporal resolution and reliability of remote sensing data (Bai et al., 2021). In recent years, researchers have combined remote sensing observation data with models to assess the impact of aerosols on vegetation productivity. Such as biogeochemical models including TEM, Biome-BGC, CENTURY, and BEPS (Amthor et al., 2001; Liu et al., 2005; Zhao et al., 2022), which can represent the dynamic changes and characteristics of biogeochemical cycles, characterize ecosystem responses to environmental changes, and provide useful support for evaluating the potential pathways through which aerosols are associated with vegetation productivity (Quillet et al., 2010). Nowadays, the effect of atmospheric aerosols on the productivity of marine vegetation has been researched, while their effect on the productivity of terrestrial vegetation has not been studied comprehensively and deeply enough. Aerosols play a pivotal role in atmospheric deposition and participate in the marine ecosystem cycle, transporting metal elements, nutrients, and other substances carried by the atmosphere above the ocean into the ocean, providing the trace elements necessary for the growth of marine plants (Westberry et al., 2023; Jickells et al., 2005). Research about the impact of aerosols on terrestrial vegetation productivity has primarily centered on temporal and spatial variation patterns and trends (Ma et al., 2022; Zhou H et al., 2024), the differences in the effects of aerosols on vegetation productivity under varying environmental conditions (Xu et al., 2023; Lin et al., 2020), and the mechanisms of direct radiation and scattered radiation on vegetation productivity (Zhou et al., 2021; Rap et al., 2018; Liu et al., 2025). For example, some scholars have observed a significant decrease in atmospheric AOD levels and a marked increase in vegetation greenness and productivity during the pandemic, as human industrial and other activities were reduced (Ranjan et al., 2022; Ranjan et al., 2020). This provides new evidence for atmospheric environmental governance. In particular, studies have indicated that the impact of aerosols on vegetation productivity is uncertain. An increase in aerosol concentration can weaken direct solar radiation, thereby inhibiting vegetation productivity. However, this process is accompanied by an increase in diffuse radiation, which allows the shaded side of vegetation leaves to receive more light and thus improves the overall light use efficiency of the canopy to some extent (Mercado et al., 2009; Zhou H et al., 2020). It is worth noting that the increase in productivity resulting from diffuse radiation may not necessarily offset the loss of productivity caused by reduced direct radiation. Moreover, elevated aerosol concentrations may simultaneously exert stress or damage on vegetation, and their positive effects generally exist only within certain concentration ranges. Moreover, aerosol deposition on leaf surfaces can also induce various physiological changes in vegetation, such as alterations in carbon uptake and transpiration rates (Ranjan et al., 2024), making it difficult to directly determine the impact of aerosols on vegetation productivity.

As one of the countries seriously affected by atmospheric aerosols, the impact of atmospheric aerosols on the productivity of vegetation in terrestrial ecosystems has become a cutting-edge topic in the current research of environmental science and ecology in China. In recent years, the implementation of a series of national policies, such as “emission reduction” and “carbon peaking,” has led to a notable decrease in the concentration of atmospheric aerosols in China. Meanwhile, vegetation productivity is also strongly regulated by climatic and anthropogenic factors, such as precipitation, temperature, and land-use change, which may be associated with aerosol dynamics and consequently influence the observed statistical relationship between AOD and vegetation productivity. Previous studies have indicated that the presence of aerosols can alter precipitation and temperature conditions, thereby increasing vapor pressure deficit (VPD) and affecting vegetation carbon sequestration and other ecosystem functions (Zhou Y et al., 2020; Salzmann, 2016) reported that global mean precipitation was more sensitive to aerosol cooling than to greenhouse gas warming, and that the aerosol-induced effect could even mask the precipitation increase associated with greenhouse gas warming (Salzmann, 2016). However, the potential differences in vegetation responses under different aerosol conditions remain insufficiently understood. This lack of knowledge hinders our understanding of the carbon cycle process in terrestrial ecosystems and the response patterns of vegetation productivity. This, in turn, affects the scientific evaluation and optimization of relevant ecological and environmental policies. Therefore, it is of great scientific significance to accurately quantify the response of vegetation productivity to aerosol dynamics and to assess the potential for ecosystem restoration under emission reduction policies, to synergize air pollution control with the goal of “carbon neutrality. Accordingly, this study used the asymmetry index (AI) combined with a time-series sliding window to quantify the long-term, high-resolution responses of vegetation productivity to aerosols, as well as the gains and losses in vegetation productivity across China. This will provide scientific references for the study of China’s regional atmospheric environmental management and the mechanism of aerosols’ perturbation to the productivity of ecosystems.

2 Study area and data

2.1 Study area

China is located in the eastern part of the Eurasian continent and on the western coast of the Pacific Ocean (73°40′-135°05′E; 18°15′-53°33′N). As a vast country spanning nine climatic zones, it provides diverse conditions for the survival of different vegetation types (Figure 1). From the coniferous forests in the northeast to the evergreen broad-leaved forests in the south, and from the desert steppes in the northwest to the tropical rainforests in the southwest, a complete spectrum of vegetation zones has been formed. Moreover, China’s diverse topography creates distinct environmental conditions for aerosol generation and accumulation. Key regions include the Loess Plateau and the Northwest Arid Zone—Asia’s major dust sources—as well as the Eastern Plains, where intensive industrial emissions dominate anthropogenic aerosol production. Additionally, enclosed terrains like the Sichuan Basin act as natural aerosol reservoirs due to their stagnant meteorological conditions. However, the promotion of clean energy and the implementation of ecological projects (e.g., the “Three-North” protection forests) have effectively reduced the aerosol concentration in recent years. This provides a broad and multi-faceted research area and conditions for exploring the response mechanism of ecosystem productivity to atmospheric aerosols.

2.2 Data sources

2.2.1 Atmospheric aerosol data

AOD characterizes the integrated aerosol load within the atmospheric column from the surface to the top of the atmosphere. Satellite remote sensing enables large-scale spatial monitoring, compensating for the sparse and uneven distribution of ground-based observations. However, satellite-retrieved AOD data are often affected by non-random missing values and high spatiotemporal variability. The AOD data used in this study are derived from the LGHAP dataset developed by Bai Kaixu et al. (https://zenodo.org/records/5655807), the first long-term, high-resolution, gap-free AOD dataset for China. The dataset integrates multisource satellite observations, ground-based measurements, and numerical simulations through multisource heterogeneous data fusion, missing information reconstruction, and deep learning techniques. Taking the Terra-MODIS AOD as the baseline, six satellite-based AOD products were bias-corrected and harmonized to a 1 km resolution using a random forest framework. Five aerosol reanalysis and numerical simulation products, including MERRA-2, were combined with ground-based AOD and meteorological visibility data to fill satellite data gaps. A spatiotemporal data-source tensor was constructed. High-order singular value decomposition (HOSVD) was used for iterative reconstruction, producing a seamless 1 km AOD product covering China from 2000 to 2020. Validation against Aerosol Robotic Network (AERONET) observations showed strong agreement, with a correlation coefficient (R) of 0.91 and a root mean square error (RMSE) of 0.21. Additional metrics such as mean absolute error (MAE), bias, and expected error (EE = ± (0.05 + 0.15 × AODsite)) were also reported. Uncertainty mainly arises from variations in satellite coverage, cloud contamination, and complex topography. Slightly higher errors were found in high-elevation or observation-sparse regions. Overall, the LGHAP dataset demonstrates high accuracy and spatial stability across China (Bai et al., 2021).

2.2.2 Aboveground carbon storage (AGC)

The data were obtained from the Tibetan Plateau Data Center (https://data.tpdc.ac.cn/), with a temporal resolution of years and a spatial resolution of 0.008333°. This dataset is based on extensive field measurements of above-ground biomass carbon (AGBC) conducted between 2011 and 2015. It was calibrated using SAR-based high-resolution maps of China’s forest aboveground biomass and integrated multiple remote sensing observations with intensive field measurements via regression and machine learning methods, yielding remote sensing imagery of China’s forest aboveground biomass carbon spanning 2002 to 2021. The results of the data-estimated changes in forest biomass carbon pools were highly correlated with the results of Liu et al. (2015) for the interannual changes in AGBC in Chinese forests (R ² = 0.65) and those of Xu et al., (2023) for the interannual changes in total biomass carbon (R ² = 0.88) (Chen et al., 2023).

2.2.3 Leaf area index (LAI)

The data were derived from MOD15A2H with a spatial resolution of 500 m. This dataset selects the available “best” pixels from all acquisitions of the Terra sensor within an 8-day period. It retrieves LAI by simulating the relationship between canopy structure and spectral reflectance using a three-dimensional radiative transfer equation, which offers more theoretical basis and higher precision compared to traditional empirical relationships. In this study, the data were processed and downloaded via the Google Earth Engine platform (https://code.earthengine.google.com/).

2.2.4 Net primary productivity (NPP)

The data is downloaded from the MOD17A3HGFv061 dataset available on the NASA Earth Science Data website (https://lpdaac.usgs.gov/). It is calculated based on the radiation use efficiency theory, incorporating remote sensing data of LAI, FPAR, and meteorological data, with respiratory consumption deducted. In addition, data imputation and quality control have been implemented to ensure the reliability of the dataset, which boasts advantages including global coverage, high spatiotemporal resolution, and high accuracy in multi-source data fusion.

2.2.5 Gross primary productivity (GPP)

The GPP data used in this study are based on the China FLUX long-term network observations and publicly available datasets, integrated with climatic, soil, and biological factor (https://doi.org/10.57760/sciencedb.o00119.00077). A random forest regression model was applied to upscale site-level AGPP to the regional scale, enabling spatially explicit simulation of the original AGPP observations. For data quality control, site observations were processed using a standardized workflow or subjected to peer review to ensure reliability and consistency. Regional AGPP data were rigorously evaluated during the model training phase, achieving an R ² of 0.85, and further assessed by comparison with existing AGPP products (e.g., GLASS, FLUXCOM, and the North American Carbon Program) in terms of spatial distribution, temporal variation, and total magnitude. Evaluation results indicate that the dataset exhibits high spatial consistency (regression slopes close to 1, R ² > 0.95) and maintains good agreement with mainstream products in temporal trends and total value (FAN et al., 2023).

2.2.6 Other supplementary data

In addition to the primary data, this study utilized four supplementary datasets to ensure its completeness. The industrial gas emissions data were sourced from the Ministry of Ecology and Environment of the People’s Republic of China (https://www.mee.gov.cn/). Precipitation data were obtained from the National Centers for Environmental Information (NCEI) under the U.S. National Oceanic and Atmospheric Administration (NOAA) (https://www.ncei.noaa.gov/). Based on daily precipitation data from all stations in China, the required national annual precipitation grid data with a spatial resolution of 1 km were interpolated using the inverse distance weighting method. Elevation data were derived from the Resource and Environmental Science Data Center, Chinese Academy of Sciences (http://www.resdc.cn), and were generated by resampling the latest SRTM V4.1 data. It adopts the WGS84 ellipsoid projection and features strong realism and high resolution. Temperature data were obtained from the National Tibetan Plateau Data Center (https://data.tpdc.ac.cn/). These data were generated through downscaling in China using the Delta spatial downscaling scheme based on the 0.5° global climate data released by CRU and the global high-resolution climate data released by WorldClim, and validated with data from 496 independent meteorological observation points, with a reliable validation result (Shouzhang, 2024).

All the above data were unified to a spatial resolution of 1 km and a geographic coordinate system (GCS-WGS-1984) using resampling and projection tools, ensuring consistent row-column numbers and resolution across datasets for the uniformity of subsequent spatial analyses (Table 1). However, these datasets have different native spatial resolutions and spatial information. Although they were resampled to a common 1-km spatial grid, differences in their original data acquisition methods, accuracy, and spatial representativeness may still result in scale-matching errors among datasets.

TABLE 1. Overview of all data used in the present study.

3 Methods

3.1 Regression analysis

Regression Analysis was used to analyze the evolution patterns of AGC, LAI, NPP, and GPP over time, the spatial change patterns, and the correlations with AOD in China during the study period (Equation 1). The results provide a basis for the subsequent analysis of vegetation productivity in response to changes in AOD.

S l o p e = n × Σ ⅈ = 1 n i × V a r i − Σ ⅈ = 1 n i × Σ ⅈ = 1 n V a r i n × Σ ⅈ = 1 n i 2 − Σ ⅈ = 1 n 2 (1)

In the formula, n represents the length of the time series during the study period, and Var _*i* denotes the value of the variable in the i-th year. When representing the long-term trend of each variable during the study period, if the Slope value is positive, the variable exhibits an increasing trend during the study period. If the value is negative, the variable exhibits a decreasing trend during the study period. The magnitude of the Slope value indicates the extent of the change. When representing the correlation between each variable and AOD, if the value is positive, the variables are positively correlated with AOD. If the value is negative, the variables are negatively correlated with AOD. The magnitude of the slope reflects the rate of change over time.

3.2 Pearson correlation coefficient

To analyze the spatial distribution characteristics of the correlation between AGC, LAI, NPP, GPP, and AOD in China during the study period, this study used the Pearson correlation coefficient (Equation 2) to fit the trend of each pixel over time in the long-term time series data, thereby revealing the spatial heterogeneity of the relationship between atmospheric aerosols and vegetation productivity.where X _*i* represents the values of AGC, LAI, NPP, and GPP; Y _*i* represents the value of AOD; X̅ and Y̅ denote the long-term sequence means of X and Y, respectively. The denominator is the product of the standard deviations of X and Y, which is used to standardize the covariance so that it is not affected by the units of the variables. The numerator is the covariance, which measures the common trend of change between X and Y.

r = C o v X , Y σ X σ Y = Σ X i − X ¯ Y i − Y ¯ Σ X i − X ¯ 2 Σ Y i − Y ¯ 2 (2)

3.3 Asymmetry index (AI)

The asymmetry index standardizes the deviation of extreme values from the mean in a set of data, decomposing the variability of the variable into positive and negative components, thereby effectively eliminating the influence of units of measurement. Additionally, by quantifying the difference between the two, the relative strength of positive and negative variability can be clearly identified, thereby precisely revealing the directionality of the response. This study used the asymmetry index (Equations 3–5) from the article of Alan K. Knapp (Knapp and Smith, 2001) and Fan Lei (Fan et al., 2024) to quantify the response mechanism of vegetation productivity to AOD changes. where Var represents the values of AGC, LAI, NPP, and GPP. Max and min represent the maximum and minimum values of the actual observations for each pixel in the multi-year time series of each vegetation productivity parameter during the study period (2002–2020). Mean represents the average value during the study period. The positive pulse indicates an increase in vegetation productivity caused by years with low AOD values, while the negative decline indicates a decrease in vegetation productivity caused by years with high AOD values. The difference between the two is the Asymmetry Index. If AI > 0, the vegetation productivity response to atmospheric aerosols is a positive asymmetric response, indicating that the increase in vegetation productivity caused by years with low AOD values during the study period exceeds the loss in vegetation productivity caused by years with high AOD values, and vegetation productivity has been effectively compensated. If AI < 0, the response of vegetation productivity to atmospheric aerosols is a negative asymmetric response, indicating that the loss in vegetation productivity caused by high AOD years during the study period is greater than the increase in vegetation productivity caused by low AOD years, and vegetation productivity has not been effectively compensated. If AI = 0, the increase and decrease in productivity caused by atmospheric aerosols during the study period offset each other, and overall productivity remains unchanged.

P o s i t i v e p u l s e = max V a r − m e a n v a r a b s m e a n V a r (3)

N e g a t i v e d e c l i n e = m e a n V a r − min V a r a b s m e a n V a r (4)

A I = P o s i t i v e p u l s e − N e g a t i v e d e c l i n e (5)

3.4 Time series sliding window

To conduct an in-depth analysis of the response, compensation, and compensation intensity trends of AGC, LAI, NPP, and GPP on AOD during the study period, this study used a time series sliding window to divide long time series data into fixed-size continuous sub-sequences for detailed analysis. The principle diagram of the time series sliding window is shown in Figure 2.

For a time series {X _*1* , X _*2* , X _*T*} of length T, the window size is set to W, meaning each window includes W consecutive years. The sliding step size is n, meaning the initial years of each window differ by n years. The first sliding window contains X _*1*, X _*2*, …, X _*w*. The second sliding window contains X _*1+n*, X _*2+n*, …, X _*w+n,* and so on, with the last sliding window containing X _*1+T-w*, X _*2+T-w*, …, X _*T*. Each window is treated as a group and undergoes quantization processing to obtain the changes in the continuous sub-time series.

This study used a 5-year sliding window with a 1-year step size for sequential calculations. Each window contains data from five consecutive years, with the starting values of each window differing by 1 year, thus gradually covering all data throughout the study period. The first sliding window, for instance, includes data from 2002 to 2006, the second, from 2003 to 2007, and so on, with the last spanning from 2016 to 2020. The use of a 5-year sliding window was mainly based on the following considerations: (1) a 5-year sliding window can effectively smooth short-term disturbances caused by interannual climatic variability while retaining sufficient temporal resolution to capture key transition periods; (2) some atmospheric pollution control policies in China were implemented over approximately 5-year periods, such as the Air Pollution Prevention and Control Action Plan; and (3) this choice is consistent with previous studies. Fan et al. (2024) also adopted a 5-year sliding window in their AI analysis in China, while Dai et al. used a 5-year sliding window in their global-scale study (Dai et al., 2025). This method effectively reduces the interference of interannual fluctuations through local calculations and captures local dynamic changes.

4 Results and analysis

4.1 Interannual variations and correlations between AOD and vegetation productivity

4.1.1 Analysis of interannual variations in AOD and vegetation productivity

During the study period, the overall AOD in China fluctuated within the range of 0.293–0.397, with a slope of −0.0034 for the fitted line against the time series, indicating a fluctuating decreasing trend of AOD over time. Specifically, in the early stage of the study (2002–2011), AOD showed a fluctuating increase from 0.324 in 2002 to 0.397 in 2011. A critical turning point occurred in 2012, after which AOD continued to decline to 0.29 (Figure 3). This significant change is likely attributed to the implementation of relevant national policies in recent years. For instance, the “12th Five-Year Plan” (2011) proposed developing clean energy, optimizing the energy structure, and reducing consumption of traditional fossil fuels like coal. This helped cut emissions of atmospheric aerosol precursors—such as sulfur dioxide, nitrogen oxides, and particulate matter—produced during combustion. In turn, it effectively lowered atmospheric aerosol concentrations. Additionally, the revised Air Pollution Prevention and Control Law of the People’s Republic of China has been in effect since 2016; it strengthened pollution control across multiple areas, including industrial pollution prevention, motor vehicle emission standards, dust pollution, and agricultural pollution. This has further advanced the progress of aerosol governance and provided an ideal research context and data support for in-depth studies on the response mechanisms of vegetation productivity to aerosol changes.

In contrast to the variation trend of AOD, AGC, LAI, NPP, and GPP showed consistent changes during the study period, exhibiting an overall continuous growth trend. AGC increased most significantly, from 1.835 kg C/m² in 2002 to 2.196 kg C/m² in 2020. This was followed by LAI and GPP, which increased by 0.154 and 0.160 kg C/m², respectively. In comparison, NPP grew relatively slowly but still increased from 0.412 kg C/m² to 0.453 kg C/m².

The spatial distribution pattern of the variation trends between AOD and vegetation productivity during the study period is shown in Figure 4. The variation trend of atmospheric aerosols exhibits distinct regional characteristics, presenting a block-like concentrated distribution. Areas with increasing AOD are relatively limited, accounting for approximately 13.39% of China’s total area, and are mainly concentrated in Northwestern China. This phenomenon may result from the combined effects of sand-dust weather and anthropogenic factors. The AOD reduction area accounts for 86.62% of the total area of the country, covering almost all regions of China except the Northwest. In particular, about 58.93% of the areas with a decreasing trend in AOD, but the trend is not obvious, are located in the Northeast, North, Northwest, and Southwest regions of China. About 26.42% of the regions with a significant decreasing trend in AOD were mainly distributed in Southeastern China, and about 1.27% of the regions with a very significant decreasing trend in AOD were distributed near the Sichuan Basin. This phenomenon may be related to its terrain surrounded by mountains on all sides, which tends to result in persistent static and stable weather conditions, characterized by low wind speed, weak pressure gradient, and stable atmospheric stratification. Such conditions are unfavorable for the diffusion of pollutants and prone to cause their accumulation near the ground. The Sichuan Basin is regarded as one of the regions with the most severe air pollution in China (Li and Xiao, 2020). Meanwhile, volumes of research have shown that precipitation leads to an extremely significant decrease in AOD (Dan et al., 2023). The extremely significant decrease in AOD in the Sichuan Basin indicates that air pollution in this region has been effectively controlled in recent years through the combined effects of natural and anthropogenic factors, which also provides abundant data support for exploring the impact of AOD changes on vegetation productivity.

The area of AGC decrease during the study period accounted for 59.27% of the national area, which is greater than the area of increase. Specifically, the areas of highly significant increase were mainly concentrated in parts of South China, accounting for 4.46% of the national area. The highly significant decrease area is distributed in localized areas of the Southwest and Northeast, accounting for about 2.51%. The largest proportion is the small decrease area, about 46.23%, mainly in East China and South China. Compared with regions with significant increases, those with highly significant increases were more centrally distributed in Eastern and Southern China. Regions with decreasing LAI accounted for 14.42%, mainly distributed in the Eastern coastal region of China, as well as small parts of Southwestern, Northwestern, and Northern China. Areas with increasing NPP accounted for 82.70% and were widely distributed. Specifically, regions with highly significant increases accounted for approximately 11.63%, which were relatively concentrated in the Northeast China Plain, North China Plain, and Shandong Hills. Areas with decreasing NPP accounted for a smaller proportion and were widely distributed in Southeastern China. The proportion of areas with increasing GPP was much higher than that with decreasing GPP, accounting for approximately 81.55%. In particular, regions with slight increases accounted for about 58.03%, while areas with significant increases were widely distributed. Regions with highly significant increases accounted for 6.46%, distributed in small parts of Xinjiang, North China, and Northwest China. Areas with decreasing GPP were relatively few, mainly distributed in parts of the Middle and Lower Reaches of the Yangtze River Plain, as well as provinces such as Fujian and Guangdong in the Southeastern coastal area.

4.1.2 Correlation analysis between AOD and various vegetation productivity parameters

In this study, regression analysis was used to analyze the correlation between the annual mean values of AOD and vegetation productivity during the study period. The results showed that AGC, LAI, NPP, and GPP exhibited a high degree of consistency in their relationship with AOD, all showing a significant negative correlation (Figure 5). The correlation coefficients were −0.61 (AGC, p < 0.01), −0.61 (LAI, p < 0.01), −0.56 (NPP, p < 0.05), and −0.49 (GPP, p < 0.05), respectively. This indicates that as atmospheric aerosols concentration increases, atmospheric transparency decreases, reducing the amount of solar radiation reaching the ground, weakening the amount of solar radiation reaching the ground, reducing the efficiency of photosynthesis, and inhibiting the growth of vegetation. Particularly, among the four parameters, AGC exhibited the largest regression slope against AOD (Slope _*AGC* = −2.210), reflecting a greater magnitude of response per unit change in AOD. Considering that vegetation productivity is also strongly influenced by temperature, precipitation, and other environmental factors, linear regression alone may not adequately distinguish the effect of AOD from those of other controlling factors. Therefore, we conducted a partial correlation analysis to examine the relationships between AOD and the four vegetation productivity indicators, with temperature and precipitation included as control variables. Among the areas showing significant correlations, 82.49% of the areas showed negative correlations between AGC and AOD, while only 17.51% showed positive correlations. For NPP, 89.57% of the areas showed negative correlations with AOD, whereas only 10.43% showed positive correlations. The proportions of areas showing negative correlations between LAI and GPP and AOD were 87.34% and 77.81%, respectively, both exceeding the proportions showing positive correlations. The partial correlation analysis further supports the robustness of the negative AOD–vegetation productivity relationships identified in this study.

4.1.3 Analysis of long-time series correlation between AOD and various vegetation productivity parameters

Based on pixel units, this study investigated the spatial correlation between AOD and vegetation productivity trends during the study period. The results showed that the long-time series correlation between vegetation productivity and AOD presented negative correlations in most regions. Specifically, approximately 64.82% of the regions showed a negative correlation in the long-time series correlation between AGC and AOD, with a relatively wide distribution. The areas where LAI, NPP, and GPP showed negative correlations with AOD in their long-time series accounted for 78.85%, 73.28%, and 70.93% respectively (Figure 6). AGC showed negative and strongly negative correlations in most areas of Southern China and the northern part of North China. The positive correlations were observed in Northeastern North China and Southwestern China, but the distribution was relatively limited. The long-time series correlations of LAI, NPP, and GPP with AOD exhibited spatial aggregation characteristics and similarities. Most parts of eastern Northeast China, Central China, Southern China, and some coastal areas showed negative correlations. Positive correlations were found in Western China, southern Northeast China, the northern part of North China, and parts of Eastern China.

4.2 Asymmetric response of vegetation productivity to AOD

This study quantified the responses of vegetation productivity parameters (AGC, LAI, NPP, and GPP) to AOD in China from 2002 to 2020 based on the asymmetry index, with the spatial distribution results shown in Figure 7. About 43.77% of the regions exhibited negative values in AGC responses, indicating that the AGC losses caused by high-AOD years during the study period were not effectively compensated in low-AOD years, showing an overall negative asymmetric response. Specifically, in high-AOD years, the high aerosol concentration enhances the extinction effect on sunlight, reducing the light available for vegetation photosynthesis at the surface and thus decreasing AGC accumulation. In low-AOD years, the relatively low aerosol concentration weakens the extinction effect, allowing sufficient light for photosynthesis to reach the surface and increasing AGC accumulation. When the reduction exceeds the increment, the AGC response to AOD presents a negative asymmetry. The areas exhibiting positive asymmetric responses for LAI, NPP, and GPP account for 64.67%, 58.87%, and 51.20%, respectively, all larger than those with negative asymmetric responses. This indicates that asymmetric responses of vegetation to AOD are widespread across China. In low-aerosol years, the losses of vegetation productivity induced by changes in atmospheric aerosols are ultimately effectively compensated. Among them, the uncompensated regions are relatively concentrated in Northeast China and eastern North China, while they are relatively scattered across South China.

To assess the robustness of the asymmetric response analysis, we additionally calculated AI using two percentile-based approaches, P10/P90 and P20/P80, instead of the original extreme values, and compared the results with those obtained using the original AI calculation method. A pixel-by-pixel comparison showed that all three methods consistently indicated a positive asymmetric response at the national scale (AI > 0), with more than 80% of the pixels showing consistent signs.

4.3 Interannual variation trend of asymmetric response of vegetation productivity to AOD

Using a 5-year time window, this study investigated the variation trends of asymmetric responses of vegetation productivity to dynamic changes in AOD from 2002 to 2020 (Figure 8). The AI of AGC and GPP were greater than 0 in all windows but showed an overall decreasing trend, indicating that during the study period, low-AOD years maintained a state of continuous compensation for vegetation AGC and GPP. However, the compensation intensity exhibited a weakening trend, as reflected by the AI (Slope _*AGC* < 0, Slope _*GPP* < 0). The AI of LAI was less than 0 only in the 2013–2017 window (indicating negative asymmetry), while AI in other windows were all greater than 0. Overall, compensation was achieved, but the compensation intensity also showed a weakening trend, as indicated by its AI (Slope _*LAI* < 0). Though there are slight fluctuations within each time window, the NPP AI remains stable at 0 overall. Notably, all vegetation productivity parameters reached their compensation peaks in the 2008–2013 window (AI _*AGC* = 0.045, AI _*LAI* = 0.011, AI _*NPP* = 0.06, AI _*GPP* = 0.015), indicating the strongest compensation intensity during this period.

In addition, considering that the window length determines the range of years included in the calculation and may affect the estimated AI values, we further conducted sensitivity tests using 3-year and 7-year sliding windows. The results showed that the overall patterns were largely consistent with those obtained using the 5-year sliding window. The finding that the strongest compensatory response occurred around 2008–2013 was also supported by both sensitivity analyses.

The spatial distribution patterns of AI for AGC, LAI, NPP, and GPP across different time windows were presented in Figure 9. By analyzing the data in each window, it was found that 47.83%, 37.87%, 42.16%, and 45.26% of the AI for each vegetation productivity parameter were positive, indicating the positive asymmetric response to changes in atmospheric aerosols. Notably, except for NPP, the mean AI of all other parameters within each window was positive. This suggests that despite the spatial extent of compensated regions being smaller than that of uncompensated ones, the compensation amount in the former was significantly higher than the loss in the latter, thus resulting in an overall positive asymmetric response. For NPP, although five windows (2005–2009, 2006–2010, 2007–2011, 2013–2017, and 2016–2020) recorded negative AI, the mean AI across all windows remained positive, so its compensation pattern was consistent with that of the other three vegetation productivity parameters. Consistent trends were observed in the spatial distribution of AI for AGC, LAI, NPP, and GPP in the 2008–2012 and 2009–2013 windows, in which, compared with earlier sliding windows, positive-value regions expanded significantly and became more concentrated, primarily in North China.

4.4 Asymmetric responses of vegetation productivity to AOD under different environmental conditions

4.4.1 Comparison of asymmetric responses in different biomes

This study categorizes vegetation into four types—forests, shrublands, grasslands, and croplands—and quantifies the AI over the study period, thereby enabling an exploration of how different vegetation communities respond to dynamic changes in AOD. The results are shown in Figure 10, which illustrates that the AI of each vegetation type (AGC, LAI, and GPP) showed positive values. As a whole, they showed a positive asymmetric response to changes in AOD, and the amount of loss was effectively compensated. In contrast, the AI of forests and croplands in NPP were negative (−0.0177 and −0.0027, respectively), and the loss was greater than the gain, showing that the loss of NPP due to changes in AOD was not effectively compensated. The AI for shrublands and grasslands were always greater than 0 (0.0893, 0.0505, 0.0144, and 0.0501; 0.1935, 0.0380, 0.0240, and 0.0577, respectively), while the values for forests were consistently lower than those for the other vegetation types. This indicates that forests compensated the least in years with low AOD values.

Quantifying the positive pulse and negative decline of AGC, LAI, NPP, and GPP for different vegetation types, as well as their correlation with AOD, provides further insight into the magnitude of vegetation productivity responses associated with AOD changes (Figures 11, 12). The results showed that grasslands exhibited the relatively strongest response to AOD changes, characterized by relatively high positive pulse (0.744 for AGC, 0.260 for NPP, and 0.298 for GPP) and negative decline (0.551 for AGC, 0.236 for NPP, and 0.241 for GPP). This relatively strong response may be partly associated with the characteristics of herbaceous vegetation, which generally has shorter life cycles, faster growth rates, and relies strongly on rapid photosynthetic carbon assimilation for biomass accumulation. These characteristics may influence how grassland ecosystems respond to changes in the light environment associated with atmospheric aerosols. Then, croplands and shrublands demonstrated a relatively strong response to changes in AOD. In contrast, forests exhibited the weakest response to AOD changes, with the lowest positive pulse (0.383 for AGC, 0.213 for LAI, 0.167 for NPP, and 0.193 for GPP) and negative decline (0.370 for AGC, 0.204 for LAI, 0.184 for NPP, and 0.177 for GPP). This may be attributed to the complex structure of forest ecosystems, which feature large biomass, long life cycles, and strong self-regulating capacity. Additionally, their canopy structure can buffer the direct impact of aerosols on photosynthesis, resulting in a weak response to short-term AOD changes.

It should be noted that different vegetation types have distinct spatial distributions across China and are associated with different climatic and environmental backgrounds. Therefore, the differences in response observed among vegetation types may reflect the combined influence of vegetation characteristics and their environmental conditions.

4.4.2 Asymmetric response of vegetation productivity under different precipitation conditions

Even in regions with consistent atmospheric aerosol concentrations, vegetation growth is affected by different environmental conditions. Precipitation plays a key role in this, directly affecting water supply and nutrient absorption in plants and thereby altering vegetation growth and productivity. Therefore, this study divided the research area into nine regions based on precipitation. Then quantified the AI, positive pulse, and negative decline of AGC, LAI, NPP, and GPP within each interval. The results demonstrated that AGC, LAI, NPP, and GPP exhibited positive asymmetric responses in areas with precipitation amounts of 600 mm or less (Figure 13). This means that the amount of vegetation productivity lost in years with high AOD values in this precipitation interval was effectively compensated for in years with low AOD values. The strongest compensation of AGC by AOD was shown by the highest AI (0.357, 0.183, 0.136, 0.064, 0.050, 0.046, and 0.021, respectively) in regions with less than 2,100 mm of precipitation. However, the amount of loss was not effectively compensated by the AI of AGC less than zero in regions with more than 2,100 mm of precipitation. It should be noted that as precipitation increases, the AI of AGC, LAI, NPP, and GPP all showed a decreasing trend, sequentially presenting negative asymmetric responses. That is, in regions with high precipitation, the phenomenon of uncompensated vegetation productivity is widespread. Additionally, with the increase in precipitation, the positive pulse and negative decline of AGC, LAI, NPP, and GPP showed a relatively consistent changing trend, with an overall decreasing tendency. This indicated that in regions with higher precipitation, the sensitivity of vegetation to AOD is lower, and the response intensity also weakens accordingly.

4.4.3 Asymmetric responses of vegetation productivity under different altitude conditions

Vegetation growing at different altitudes within the same region exhibits distinct growth patterns and responses to AOD as they adapt to different growing environments. Accordingly, we divided the study area into nine regions based on precipitation and quantified the AI, positive pulse, and negative decline of AGC, LAI, NPP, and GPP in each interval. The result showed that, except for NPP in the 0–1,000 m altitude range, vegetation productivity overall exhibited positive asymmetric responses across different elevation intervals (Figure 14). This indicates that the reductions in vegetation productivity caused by high-AOD years were effectively compensated in low-AOD years in regions at different altitudes. In particular, AOD showed stronger overall compensation intensity for AGC compared to others, characterized by higher AI, with the peak compensation observed at altitudes above 4500 m. In contrast, the responses of LAI, NPP, and GPP were relatively stable. Correspondingly, AGC displayed higher positive pulse and negative decline, showing the highest sensitivity to AOD changes. The positive pulse and negative decline of LAI and GPP exhibited similar variation trends across different altitudes, peaking at 1,000–1,500 m and then gradually decreasing. NPP differed in that its positive pulse and negative decline showed a continuous increasing trend at altitudes above 2500 m.

4.4.4 Asymmetric responses of vegetation productivity under different temperature conditions

To clarify the responses of vegetation productivity to atmospheric aerosol changes across varying temperature environments, seven research intervals were divided based on temperature gradients in this study, and the AI, positive pulse, and negative decline of AGC, LAI, NPP, and GPP within each temperature gradient were quantified. In regions with an annual average temperature below 0 °C, the AI of AGC, LAI, NPP, and GPP were all positive, indicating positive asymmetric responses to AOD changes (Figure 15). This suggests that vegetation productivity in areas with lower annual average temperatures has been effectively compensated. With increasing temperature, the AI of vegetation productivity showed an overall decreasing trend, reflecting a weakening of compensation intensity as temperature rises. When temperatures exceeded 10 °C, the compensation intensity of AOD for NPP and GPP both weakened, and even failed to achieve compensation (e.g., at 20 °C–25 °C, AI _*NPP* = −0.002, AI _*GPP* = −0.009; above 25 °C, AI _*NPP* = −0.007, AI _*GPP* = −0.007). This may be attributed to elevated temperatures stimulating increased plant respiration, which consumes more organic matter produced by photosynthesis. Additionally, high temperatures can intensify plant transpiration, leading to excessive water loss that impairs normal cellular physiological functions and thereby affects vegetation productivity. The study revealed that the positive pulse and negative decline of LAI, NPP, and GPP all peaked at 5 °C–10 °C, indicating that vegetation within this temperature range is most sensitive to atmospheric aerosol changes and exhibits the strongest response capacity. This may be due to the high activity of enzymes involved in photosynthesis (e.g., carboxylase) within 5 °C–10 °C, which efficiently catalyze reactions such as carbon dioxide fixation. Reduced atmospheric aerosol concentrations also provide sufficient light, enabling plants to fully utilize light energy for organic matter synthesis. Furthermore, within this temperature interval, plant respiration intensity is moderate and does not excessively consume organic matter produced by photosynthesis. With further increases in temperature, the positive pulse and negative decline of vegetation showed an overall decreasing trend, indicating that the response intensity of vegetation productivity to AOD changes is reduced in high-temperature regions.

5 Discussion

5.1 Analysis of asymmetric response of vegetation productivity to changes in aerosol dynamics

The results of the above studies indicate that the asymmetric response of vegetation productivity to AOD changes was widespread across China during 2002–2020, with positive asymmetric responses dominating (Figure 10). This phenomenon may be attributed to the fact that increases in vegetation productivity during low-AOD years compensate for the losses incurred during high-AOD years. In low-AOD years, vegetation photosynthesis is enhanced by sufficient sunlight for the photoreaction phase, which directly promotes productivity. Additionally, low atmospheric aerosol concentrations are typically accompanied by higher temperatures and reduced atmospheric pollutants. Within certain thresholds, these conditions can enhance photosynthetic efficiency and the activity of related enzymes, thereby further boosting vegetation productivity. Meanwhile, in years with high aerosol loading, the reduction in direct radiation is accompanied by an increase in diffuse radiation, which allows the backlit leaves in different canopy layers to receive more light. However, the enhancing effect of diffuse radiation is limited; when it is insufficient to offset the negative impact of reduced direct radiation on vegetation productivity, a negative asymmetric response occurs. In addition, changes in AOD may alter environmental conditions such as precipitation and vapor pressure deficit (VPD), thereby affecting vegetation water status and stomatal regulation and, in turn, contributing to variations in vegetation productivity. These aerosol-associated environmental changes may act together with changes in radiation conditions, ultimately contributing to the positive asymmetric response of vegetation productivity. Recent studies have confirmed that implementing policies such as the “Air Cleanup Initiative” and achieving carbon neutrality in China have contributed to the recovery of terrestrial productivity (Zhou Z et al., 2024; Li et al., 2024).

5.2 Comparison of asymmetric responses across different biomes

Grasslands exhibited a relatively strong response to changes in AOD (Figures 11, 12). Compared with shrubs, crops, and forests, grasses have shallower, less fibrous root systems and less developed stem mechanical tissues—which may influence their responses to environmental variability. Additionally, grassland ecosystems have more homogeneous community structures, making them more prone to impacts from anthropogenic activities such as grazing and reclamation, as well as environmental changes. When other environmental conditions are suitable, light becomes a key factor determining vegetation growth status and productivity. On the one hand, atmospheric aerosols—an important factor affecting direct and scattered solar radiation—may directly influence light acquisition in grass communities, which have simple structures and uniform canopies. On the other hand, vegetation light acquisition is primarily reflected in light interception capacity, light energy utilization efficiency, and light tolerance characteristics (Zhi-Hua et al., 2001; Cao et al., 2025; Jin et al., 2021). At the same time, herbaceous plants generally have lower light saturation points than some woody plants, meaning that they reach photosynthetic saturation under high-light conditions at relatively lower irradiance levels. Differences in stomatal regulation and light-damage repair capacity may further influence their ability to adjust to changes in the light environment. Taken together, these structural and functional characteristics partly contribute to the relatively strong AOD-associated response observed in grasslands in this study. In contrast, forest ecosystems exhibit the relatively weakest response to changes in aerosol concentrations, primarily due to their distinct structural and ecological characteristics. Their complex canopy structure directly determines the distribution of solar radiation within canopies at different heights, influencing the interception, absorption, and transmission of average daily photosynthetically active radiation by trees at various layers, which in turn shapes the productivity of forest ecosystems (Satapathy and Dutta, 2025). Additionally, forests possess large biomass stocks and long life cycles, which may provide greater buffering capacity against short-term environmental fluctuations.

5.3 Analysis of asymmetric response trends

Although the research results indicate that the loss of vegetation productivity in China due to changes in AOD was effectively compensated for overall during the study period (Figure 10), a 5-year sliding window analysis shows that the AI of AGC, LAI, and GPP decreased overall. This suggests that the positive asymmetric response gradually shifted to the negative asymmetric response of vegetation to AOD changes during the study period. The phenomenon may be due to the frequent occurrence of extreme drought events in recent years. These events have led to temperature and precipitation conditions that jointly inhibit plant photosynthesis, enhance respiration, and negatively impact vegetation productivity, greenness, and carbon cycling (Deng et al., 2021). In addition, it may also be attributed to plants adjusting their morphological characteristics and biomass allocation to adapt to the low-light environment caused by atmospheric aerosols, i.e., phenotypic plasticity (Che et al., 2024). For example, numerous studies have shown that vegetation increases its specific leaf area under low-light conditions, accompanied by a decrease in the light saturation point (LSP) and light compensation point (LCP). This allows vegetation to improve its ability to utilize weak light and reduce respiratory consumption (Zhang et al., 2011). Even when light intensity increases in the later stage, plants still maintain a relatively low saturation point for a certain period, resulting in an overall decrease in light use efficiency and productivity. Notably, vegetation productivity reached a compensation peak during 2008–2013, characterized by the highest AI, followed by a downward trend. This may be related to a series of ecological protection policies implemented in China. For instance, studies have shown that since the vigorous implementation of the Grain for Green Project (converting farmland to forests and grasslands), NPP, GPP, and LAI in China have all shown an increasing trend (Xie et al., 2023; Ma et al., 2024).

5.4 Influencing factors of asymmetric responses

The study finds that as precipitation increases, the AI of vegetation productivity gradually shifts from positive to negative (Figure 13), indicating that precipitation conditions are a key factor affecting the asymmetric response of vegetation productivity. When precipitation is within 900 mm, vegetation productivity is generally compensated. However, once precipitation exceeds a certain threshold, excessive water causes plant roots to undergo anaerobic respiration, reducing the efficiency of nutrient and water absorption. Coupled with increased cloud cover and reduced sunlight, these factors jointly inhibit vegetation growth (Li et al., 2024).

When the altitude is within 1,000 m, only the AI of NPP is negative, showing a negative asymmetric response to AOD changes. With the increase in altitude, the AI of AGC, LAI, NPP, and GPP all become positive. AGC responds most strongly to atmospheric aerosols, characterized by a relatively high AI (Figure 14). In contrast, the changes in LAI, NPP, and GPP values fluctuate slightly. High-altitude mountainous areas are often accompanied by low temperatures, high precipitation, strong sunlight, and more diverse ecosystems, which provide favorable conditions for vegetation growth. Therefore, compared with vegetation in low-altitude areas, vegetation in high-altitude areas generally shows a positive asymmetric response, and their productivity is effectively compensated (Hongli et al., 2009).

In regions where the annual temperature is below 15 °C, vegetation productivity generally shows a positive asymmetric response, indicating that the losses during the study period have been effectively compensated. When the temperature exceeds 15 °C, the compensation effect generally tends to weaken. This may be related to the optimal temperature for the enzymes involved in vegetation photosynthesis. Meanwhile, temperature directly affects physiological and biochemical processes such as plant respiration, cell division, and material transport. Studies have shown that the optimal temperature for vegetation growth exhibits significant regional differences. It is regulated by various environmental factors, including soil moisture, vegetation phenology, and stomatal conductance. The optimal temperature for tropical forest ecosystems is close to 30 °C, while that for temperate deciduous forests, coniferous forests, and grasslands is relatively lower, at approximately 25 °C, 20 °C, and 13 °C, respectively. When the temperature exceeds the optimal range, it will affect processes such as plant metabolism and regulation, leading to physiological decline, growth inhibition, and even death of plants (Huang et al., 2019).

Building on these findings, we further conducted interaction analyses to examine whether precipitation, temperature, elevation, vegetation type, and policy period modify the statistical relationship between AOD and GPP. For the environmental factors, the coefficients of the interaction terms between AOD and precipitation, temperature, and elevation were −0.232, 0.208, and 0.104, respectively, and all were statistically significant (p < 0.001). Vegetation type also showed a significant moderating effect on the AOD–GPP relationship. In addition, considering the distinct shift in AOD around 2012 during the study period and the implementation of atmospheric pollution control policies in China, we divided the study period into two periods, 2002–2011 and 2012–2020, and constructed a policy-period variable accordingly. The interaction term for policy period was also statistically significant (β = 0.098, p < 0.001). These results indicate that the statistical relationship between AOD and GPP varies with precipitation, temperature, elevation, vegetation type, and policy period. These findings provide statistical evidence for differences in the AOD–vegetation productivity relationship under different environmental and contextual conditions.

As one of the most uncertain factors in global climate change, atmospheric aerosols have been identified as a major driver of cloud water variability, with effects that may exceed those of meteorological factors. Changes in aerosol dynamics can significantly influence regional and global temperature and precipitation processes through both direct and indirect effects. These interactions are further complicated by differences in vegetation types, phenological conditions, and climatic zones. In this study, we used the asymmetry index to quantify the compensatory responses of vegetation productivity across different vegetation types and environmental gradients, thereby revealing how vegetation productivity responds to changes in atmospheric aerosols. Our findings provide a new approach and reference for assessing the relationship between atmospheric aerosols and vegetation productivity.

6 Implications

The findings of this study have implications for air quality and terrestrial ecosystem management in China. The asymmetric responses of vegetation productivity to AOD changes indicate that the magnitude and direction of vegetation responses to aerosols may vary under different environmental conditions. Although vegetation productivity has generally remained in a state of compensation during the study period, the compensatory effect has shown a clear weakening trend in recent years. This suggests that, against the backdrop of sustained emission-reduction policies and declining aerosol concentrations, other environmental factors—such as temperature and precipitation—are also jointly influencing changes in vegetation productivity. Therefore, while continuing to promote air-pollution reduction, attention should also be given to the concurrent changes in these environmental factors and their potential roles in modifying the aerosol–vegetation relationship. Meanwhile, further research is needed to elucidate the balance between direct and diffuse radiation induced by aerosols, as well as its influence on light partitioning within the canopy, because this balance directly determines the light use efficiency of the vegetation canopy and largely shapes the response of vegetation productivity to aerosol changes. This can help provide a more comprehensive assessment of ecosystem adaptation and recovery under changing environmental conditions. It may also improve our understanding of ecosystem productivity dynamics under improving air quality conditions.

Furthermore, the analysis of different vegetation types showed that grasslands and croplands exhibited relatively stronger responses to AOD changes, whereas forests showed weaker responses. This suggests that different ecosystems should be considered separately when assessing the effects of aerosol changes and other environmental conditions on vegetation productivity. For ecosystems such as grasslands and croplands, greater attention should be paid to their responses to changes in atmospheric and environmental conditions. Their potential changes in productivity should also be considered in assessments of air-quality improvement and ecological restoration. This may provide a useful reference for differentiated ecosystem management.

7 Limitations

This study revealed the asymmetric responses of vegetation productivity to atmospheric aerosol changes at the national scale. However, several limitations should be acknowledged.

First, this study was mainly based on remote-sensing products and statistical association analyses to characterize the statistical relationships between AOD changes and vegetation productivity and their asymmetric patterns. The AOD and vegetation productivity products also contain uncertainties, which may arise from satellite retrieval errors, cloud contamination, complex topography, and differences in data-generation methods (Ranjan et al., 2020). Although all datasets were resampled to a common 1 km spatial grid, their original observation scales, accuracy, and spatial representativeness remained different. These differences may affect the characterization of relationships among variables at the pixel scale. In addition, some datasets used in this study were derived from model-based estimates. Although these datasets have been validated against ground observations and generally show acceptable accuracy, their inherent uncertainties may still affect the analysis results.

Second, vegetation productivity is influenced by various natural and anthropogenic factors. For example, increasing atmospheric CO₂ concentrations may promote photosynthesis and vegetation productivity; nitrogen deposition may affect vegetation growth by increasing nitrogen availability; irrigation can improve soil water conditions; and land-use change and ecological restoration may alter vegetation cover, community structure, and productivity. These factors may occur simultaneously with changes in AOD, making it difficult to fully separate their individual contributions within the present framework. In addition, the manuscript mainly characterized spatial differences in asymmetric responses by examining the distribution of AI across different precipitation, temperature, and elevation gradients. However, these environmental factors are not spatially independent at the national scale, and changes along one environmental gradient are often accompanied by concurrent changes in other environmental conditions. Therefore, differences in AI observed across individual precipitation, temperature, or elevation gradients may be partly influenced by the combined effects of other correlated environmental factors. Such spatial covariation among environmental factors may increase the uncertainty of the gradient-based analysis and represents a limitation of the present study.

Third, vegetation and environmental gradients exhibit strong spatial continuity, and neighboring pixels often share similar environmental conditions. Therefore, the pixel-level analyses in this study may be affected by spatial autocorrelation. Although the calculation of AI does not involve hypothesis testing and spatial autocorrelation therefore does not directly affect its statistical significance, it may still influence the rigor and robustness of the interpretation of spatial distribution patterns. Future studies could incorporate spatial statistical models, such as spatial error models or spatial econometric models, to further quantify the potential influence of spatial autocorrelation on the results. In addition, the use of higher-resolution remote-sensing data (e.g., 10–30 m) could help reduce uncertainties associated with mixed pixels.

Finally, this study mainly revealed the asymmetric responses of vegetation productivity to changes in AOD, while there is still room to further investigate the specific processes through which aerosols affect vegetation productivity. For example, the changes in direct and diffuse radiation induced by aerosols and their relative contributions were not quantitatively distinguished in detail. Meanwhile, aerosols may also affect vegetation productivity indirectly by altering temperature, precipitation, and other environmental conditions, while these potential pathways were not fully characterized in the present study. Future research could integrate radiation observations, ecosystem process-based models, and additional environmental variables to further elucidate the complex interactions among aerosols, environmental conditions, and vegetation.

8 Future perspectives

Based on the findings and limitations of this study, future research could further investigate the processes through which aerosols affect vegetation productivity by combining multi-source observations with modelling approaches. Process-based models and causal analysis could be combined with remote-sensing observations. This would help distinguish the effects of aerosols on direct and diffuse radiation. It would also allow their interactions with temperature, precipitation, and soil moisture to be examined.

Flux-tower observations could also be used to validate vegetation productivity responses derived from remote sensing and modelling approaches. In addition, future studies could further consider anthropogenic factors such as land-use change and ecological restoration. More detailed process-based analyses could also be conducted for different ecosystem types. These efforts may help clarify the processes underlying the statistical relationships between aerosol changes and vegetation productivity. They may also provide additional evidence for understanding how different ecosystems respond to aerosol variations.

9 Conclusion

This study investigated the asymmetric responses of vegetation productivity to atmospheric aerosol dynamics in China over 19 years using the asymmetry index. The results show consistent spatial distribution characteristics in the responses of the four vegetation productivity parameters to atmospheric aerosols. Negative asymmetric responses were mostly observed in Northeastern and Northern China, while positive asymmetric responses were predominant in Southern and Central China. During the study period, vegetation productivity in China generally exhibited a positive asymmetric response to changes in atmospheric aerosols, indicating that the productivity gains associated with low-AOD conditions were generally greater than the productivity losses associated with high-AOD conditions. However, a detailed analysis of AI using a 5-year sliding window revealed a weakening trend in compensation intensity. The compensation intensity of all four vegetation productivity parameters peaked around 2008–2013, likely due to China’s ecological protection and emission reduction policies implemented around 2008. Studies on different vegetation types indicated that grasslands and farmlands responded more strongly to changes in atmospheric aerosols, while the NPP and GPP of forests showed no compensation effect, and the response of croplands was insignificant. The AI of vegetation productivity was generally higher under low precipitation conditions, with compensation observed in most precipitation ranges. As precipitation increased, the compensation intensity weakened or disappeared entirely. In terms of elevation, non-compensation only occurred in low-altitude areas. Vegetation in regions with an annual average temperature below 15 °C exhibited a positive asymmetric response and effectively compensated for productivity losses. However, due to the influence of multiple factors, the positive asymmetric response gradually shifted to a negative asymmetric response as the temperature gradient increased.

Generative AI statement

The author(s) declared that generative AI was not used in the creation of this manuscript.

Any alternative text (alt text) provided alongside figures in this article has been generated by Frontiers with the support of artificial intelligence and reasonable efforts have been made to ensure accuracy, including review by the authors wherever possible. If you identify any issues, please contact us.

Publisher’s note

All claims expressed in this article are solely those of the authors and do not necessarily represent those of their affiliated organizations, or those of the publisher, the editors and the reviewers. Any product that may be evaluated in this article, or claim that may be made by its manufacturer, is not guaranteed or endorsed by the publisher.

Footnotes

Edited by: Yu Feng, Eastern Institute of Technology, China

Reviewed by: Avinash Kumar Ranjan, Indian Institute of Technology Bombay, India

Samuel Chukwujindu Nwokolo, University of Calabar, Nigeria