Introduction

Achieving carbon neutrality and the Paris Agreement’s climate target necessitates substantial reductions in fossil fuel use, leading to a concomitant decline in anthropogenic aerosol emissions1. Aerosol decline not only benefits air quality and public health but also has profound impacts on terrestrial ecosystems through multiple pathways2,3,4. First, aerosol decline increases total solar radiation while reducing its diffuse fraction, creating competing effects on plant photosynthesis. While higher total radiation improves light availability, the concurrent reduction in diffuse radiation diminishes light-use efficiency at the canopy level by limiting light utilization in shaded leaves5,6,7. Second, aerosol decline can either enhance or suppress photosynthesis by altering climate conditions such as air temperature and precipitation8,9,10. Aerosol decline-induced climate changes even exacerbate natural disasters such as wildfires, leading to direct damage to vegetation11. Furthermore, photosynthesis is also susceptible to aerosol deposition12 and chemical processes13. These physiological and structural changes in vegetation can modify surface biophysical properties (e.g., albedo), which regulate land–atmosphere energy and water exchanges and ultimately influence climate. However, exact biophysical climate feedback from vegetation responses to aerosol decline remains largely unknown. This knowledge gap hinders a comprehensive understanding of the interactions among aerosols, vegetation, and climate.

Although biophysical climate feedback from aerosol-driven vegetation changes remains unknown, extensive studies on carbon dioxide (CO2)–vegetation interactions provide useful insights. Elevated atmospheric CO2 levels increase vegetation photosynthesis by accelerating the carboxylation rate14, promoting leaf area expansion. This expansion reduces surface albedo while increasing aerodynamic roughness and evapotranspiration efficiency, yielding competing effects on the local climate15. In general, albedo-driven warming dominates in boreal regions, whereas evapotranspiration-driven cooling prevails elsewhere16,17,18. Meanwhile, rising atmospheric CO2 concentrations enhance vegetation water-use efficiency by inducing partial stomatal closure, particularly over water-limited regions19,20,21. This vegetation physiological response reduces the amount of water loss via transpiration, thereby inhibiting evaporative cooling and resulting in local warming22,23,24. In addition to these local climate effects, the vegetation response to rising CO2 concentrations can even modulate large-scale atmospheric circulations, yielding climate effects that extend from regional to global scales25,26,27. Given different vegetation behaviors in response to changing CO2 and aerosol levels, their consequent effects on climate may differ substantially. Therefore, it is imperative to explicitly disentangle biophysical climate feedback from aerosol-driven vegetation changes and its underlying mechanisms.

This scientific question is also essential for informing climate and air quality policies. While aerosol decline is known to accelerate global warming28,29,30, a growing number of studies have shown that it can also boost CO2 uptake by vegetation in specific regions, such as China31,32. These findings suggest that aerosol decline may strengthen vegetation’s climate mitigation potential, potentially offsetting the radiative warming effect from aerosol decline. However, these studies focused exclusively on biochemical feedback (i.e., the carbon cycle) from vegetation responses to aerosol decline while neglecting biophysical feedback. Unlike biochemical feedback that commonly operates at a global scale, biophysical feedback has more immediate and localized climatic consequences, particularly in populated regions near vegetated areas33,34,35. Such oversight may undermine accurate assessment of the local climate mitigation potential of vegetation-based solutions, such as afforestation or reforestation, under low-aerosol conditions. Furthermore, the role of vegetation biophysical feedback in aerosol climate effects remains unclear, yet it is critical for accurately projecting the climate impacts of air pollution control. In summary, ignoring vegetation biophysical feedback risks compromising the effectiveness of climate and air quality policies or, worse, producing unintended counterproductive outcomes.

In this study, we investigate biophysical temperature feedback from vegetation responses to aerosol decline at the global scale. Specifically, we aim to address the following three key questions: (1) What are the spatial and seasonal patterns of feedback? (2) What mechanisms drive such feedback? (3) What is the relative importance of biophysical versus biochemical feedback in terms of climate mitigation? To address the first and second questions, we use multiple state-of-the-art Earth system models that can simulate vegetation responses to changes in aerosol emissions. Notably, isolating vegetation feedback on temperature is challenging, as it is readily obscured by the radiative warming effect of aerosol decline itself. To overcome this issue, we use an empirical model to estimate vegetation feedback and elucidate its underlying mechanisms. To isolate the aerosol effect, we use simulations under multiple air pollutant emission scenarios from the Coupled Model Intercomparison Project Phase 6 (CMIP6) and CMIP6-endorsed Model Intercomparison Projects. To address the third question, we quantify the impact of aerosol decline on terrestrial carbon fluxes and convert this carbon effect to an equivalent temperature metric, enabling its direct comparison with our estimated vegetation biophysical feedback.

Results

Impacts of idealized aerosol decline on vegetation leaf area

We begin by analysing vegetation biophysical feedback under an idealized scenario with complete removal of anthropogenic aerosol emissions. To this end, we use two fully coupled simulations: the historical simulation from the CMIP636 and the hist-piNTCF simulation from the Aerosol Chemistry Model Intercomparison Project37 (AerChemMIP; see Methods). The historical simulation spans 1850–2014 and incorporates all natural and anthropogenic forcings. The hist-piNTCF simulation is identical to the historical simulation except that anthropogenic aerosol and ozone precursor emissions are maintained at preindustrial levels. Differences between these two simulations (hist-piNTCF minus historical) isolate the effect of aerosol decline. Notably, while these two simulations differ in ozone emissions, the ozone effect is minimal, which is discussed later. We focus on the last 30 years (i.e., 1985–2014) because the emission difference between the two simulations reaches a maximum during this period (Supplementary Fig. S1). Five models (BCC-ESM1, CNRM-ESM2-1, NorESM2-LM, TaiESM1, and UKESM1-0-LL) provide all necessary outputs, and their results are averaged to produce the multimodel mean for analysis. We restricted the analysis to vegetated areas where the leaf area index (LAI) exceeds 0.1 m2·m−2. The results are considered robust when at least four out of the five models agree on the sign of the multimodel mean, as indicated by the black dots in the figures.

As a result of the idealized emission reductions (Supplementary Figs. S1 and S2), aerosol loading, as indicated by the aerosol optical depth at 550 nm, decreases substantially over continents, particularly in heavily polluted regions such as eastern China, India, Europe and the eastern United States (Fig. 1a). This aerosol decline boosts vegetation growth, as reflected by LAI changes (Fig. 1b). On average, annual LAI increases by 0.14 ± 0.08 m2·m-2 (hereafter, multimodel mean ± intermodel standard deviation) in boreal regions (see Fig. 1F for the climate zone) and 0.04 ± 0.03 m2·m−2 in temperate regions. This LAI enhancement is mainly due to increased total solar radiation and higher temperature following aerosol decline (Supplementary Figs. S3a, S3c, S4a and S4b), which alleviates light and thermal constraints on vegetation growth4,8,9. Additionally, land surface warming following aerosol decline provides more energy for evaporation and convection, leading to increased precipitation (Supplementary Figs. S3d, S5a, S5b), which further facilitates LAI expansion in water-limited regions.

Fig. 1: Annual effects of idealized aerosol emission reductions.
Fig. 1: Annual effects of idealized aerosol emission reductions.
Full size image

a Changes in the aerosol optical depth at 550 nm (unitless). b Impacts of aerosol decline on the leaf area index (LAI; unit: m2·m−2). c Baseline temperature sensitivity to LAI (dT/dLAI; unit: °C·m−2·m2) in the historical simulation. d, e Impacts of aerosol decline on dT/dLAI (Δ(dT/dLAI); d), with the impacts categorized into four classes (e): stronger warming (dT/dLAI >0 and Δ(dT/dLAI)>0), weaker warming (dT/dLAI >0 and Δ(dT/dLAI)<0), weaker cooling (dT/dLAI <0 and Δ(dT/dLAI)>0) and stronger cooling (dT/dLAI <0 and Δ(dT/dLAI)<0). f Map of boreal, temperate and tropical regions determined by the Köppen-Geiger climate classification. In (a–e), shading denotes the multimodel mean. In (a, b, d, e), black dots denote where at least four out of the five models agree on the sign of the multimodel mean. In (b–d), imbedded bar plots show regional averages across boreal (blue), temperate (gray), and tropical (orange) vegetated areas, with bars and error bars representing the multimodel mean and intermodel standard deviation, respectively. The results are derived from the comparison between the historical and hist-piNTCF simulations (hist-piNTCF minus historical) over 1985-2014.

In contrast, annual LAI decreases by 0.08 ± 0.04 m2·m−2 in tropical regions, particularly in the Amazon basin and Central Africa. This LAI reduction occurs mainly because the tropical temperature is already at photosynthetic optimum levels, and additional warming from aerosol decline (Supplementary Figs. S3a and S4c) inhibits vegetation growth4,8,9. Moreover, aerosol decline-induced reductions in diffuse radiation (Supplementary Figs. S3b and S4c) may also contribute to the LAI reduction. This pattern of LAI responses to aerosol decline is robust across models and aligns well with previous studies reporting opposite responses of vegetation productivity to aerosols at low versus high latitudes4.

The strength of LAI responses also varies with season, particularly in boreal and temperate regions (Fig. 2a–c), depending on the combined radiative and climate effects of aerosol decline (Supplementary Figs. S4 and S5). For example, in boreal regions, the LAI response is more pronounced during summer, primarily driven by larger increases in total solar radiation and precipitation.

Fig. 2: Seasonal effects of idealized aerosol emission reductions.
Fig. 2: Seasonal effects of idealized aerosol emission reductions.
Full size image

a–c Baseline leaf area index (LAI; unit: m2·m-2; blue line), and its response to aerosol decline (red line). d–f Baseline temperature sensitivity to LAI (dT/dLAI; unit: °C·m-2·m2; blue line), and its response to aerosol decline (red line). g–i Biophysical feedback of aerosol decline-induced vegetation changes on the near-surface air temperature (unit: °C; black line), along with individual contributions from LAI responses (blue line) and dT/dLAI responses (orange line). The results are shown for the boreal (a, d, g), temperate (b, e, h) and tropical (c, f, i) regions. In each panel, all the results are regional averages across vegetated areas, with the line and shading representing the multimodel mean and the intermodel standard deviation, respectively. The baseline is derived from the historical simulation, and the response is derived from the comparison between the historical and hist-piNTCF simulations (hist-piNTCF minus historical) over 1985-2014.

Impacts of idealized aerosol decline on the temperature sensitivity to leaf area changes

Vegetation biophysical feedback on temperature depends not only on LAI changes but also on the temperature sensitivity to LAI changes (dT/dLAI). dT/dLAI quantifies the local response of the near-surface air temperature to each unit increase in LAI. We estimate dT/dLAI across vegetated areas using a well-validated empirical model, which is a function of surface solar radiation, evapotranspiration and snow cover38 (see Methods). This model is superior in its minimal input requirements while effectively accounting for vegetation feedback on local temperature through biophysical processes (e.g., albedo and evapotranspiration). The sign of dT/dLAI reflects the balance between warming from reduced albedo and cooling from enhanced evapotranspiration following LAI enhancement15,16,17,18.

The baseline annual dT/dLAI in the historical simulation exhibits distinct latitudinal variability, with widespread positive values (indicating warming per unit increase in LAI) in boreal regions but negative values (indicating cooling per unit increase in LAI) in temperate and tropical regions (Fig. 1c). Positive dT/dLAI in boreal regions is due to enhanced albedo-driven warming during snow-covered seasons (Fig. 2d), which dominates the annual temperature response. More specifically, LAI enhancement increases snow cover masking, creating larger-than-normal albedo reductions that ultimately produce net warming16,17,18,39. In contrast, in temperate and tropical regions, the dominant role of evapotranspiration-driven cooling16,17,18,39 drives dT/dLAI to be negative throughout the year (Fig. 2e and f). Additionally, the magnitude of dT/dLAI is, overall, larger in low-LAI regions than in high-LAI regions, probably because the evapotranspiration sensitivity to LAI saturates at high LAI levels16,18,39. In other words, the evaporative cooling per unit increase in LAI diminishes as LAI increases, leading to a less negative dT/dLAI value. In short, the global pattern of our estimated dT/dLAI aligns well with previous findings16,17,18,39.

Aerosol decline causes robust reductions in annual dT/dLAI across most vegetated areas, with exceptions in a few regions, such as southern Australia, Argentina, and eastern Central Africa (Fig. 1d). However, their underlying mechanisms vary by region, depending on how aerosol decline modulates the albedo and evapotranspiration effects of LAI enhancement under varying climate conditions. The largest reduction in dT/dLAI (-0.0050 ± 0.0028 °C·m−2·m2) occurs in boreal regions, indicating weaker warming per unit increase in LAI. This is mainly due to the reduced snow cover following aerosol decline (Supplementary Fig. S5), which alleviates the albedo-driven warming effect during the cold season (Fig. 2d). The dT/dLAI reductions in temperate (-0.0019 ± 0.0035 °C·m−2·m2) and tropical (-0.0026 ± 0.0025 °C·m−2·m2) regions indicate greater cooling per unit increase in LAI. This mainly results from higher solar radiation (Supplementary Fig. S4b, S4c) and evapotranspiration (Supplementary Fig. S5e and f) following aerosol decline, both of which enhance the evapotranspiration-driven cooling effect (Fig. 2e, f). In addition, for tropical regions, decreased LAI enhances the evapotranspiration sensitivity to LAI, further reducing dT/dLAI (i.e., more negative). In summary, the dT/dLAI reduction implies that each unit increase in LAI would produce weaker warming in boreal regions but stronger cooling in temperate and tropical regions (Fig. 1e).

Biophysical temperature feedback from vegetation responses to idealized aerosol decline

We next examine vegetation biophysical feedback on the local near-surface air temperature due to aerosol decline, which is estimated as the difference in the product of LAI and dT/dLAI between the historical and hist-piNTCF simulations (see Methods). Furthermore, we decompose this feedback into separate contributions from LAI responses and dT/dLAI responses by perturbing each factor individually with the other fixed (see Methods).

At the annual scale, biophysical temperature feedback is robust in most vegetated areas and shows distinct spatial heterogeneity (Fig. 3a). Cooling effects occur in East Asia, northern Eurasia, North America, and the Amazon basin, whereas warming effects can be found in Southeast Asia, Central Africa, and Colombia. This feedback is due to synergic or competing effects from responses of LAI and dT/dLAI to aerosol decline (Fig. 3b, c). For East Asia, both LAI and dT/dLAI responses yield cooling, which acts synergistically to produce stronger cooling (up to <−0.03 °C). For the Amazon basin, northern Eurasia, and North America, cooling from dT/dLAI responses outweighs warming from LAI responses, leading to net cooling (−0.03 to 0 °C). In contrast, for Central Africa, warming from LAI responses exceeds cooling from dT/dLAI responses, leading to net warming (0 to 0.02 °C). On average, vegetation feedback causes local cooling in boreal (−0.0083 ± 0.0070 °C) and temperate (−0.0036 ± 0.0017 °C) regions but relatively weak local warming in tropical regions (0.0007 ± 0.0011 °C). Notably, these regional mean values are small because of the strong spatial heterogeneity in temperature feedback, which causes opposing outcomes to cancel each other out. Globally, dT/dLAI responses dominate feedback in ~72% of vegetated areas, whereas LAI responses are dominant in the remaining ~28% (Fig. 3d).

Fig. 3: Annual biophysical feedback from vegetation responses to idealized aerosol emission reductions.
Fig. 3: Annual biophysical feedback from vegetation responses to idealized aerosol emission reductions.
Full size image

a Feedback on the near-surface air temperature (unit: °C). Feedback attributable to (unit: °C) LAI responses (b) and dT/dLAI responses (c), respectively. d The dominant driver of feedback, with blue denoting LAI responses and orange denoting dT/dLAI responses. In each panel, shading denotes the multimodel mean, and black dots denote where at least four out of the five models agree on the sign of the multimodel mean. In (a–c), embedded bar plots show regional averages across boreal (blue), temperate (gray), and tropical (orange) vegetated areas, with bars and error bars representing the multimodel mean and intermodel standard deviation, respectively. The results are derived from the comparison between the historical and hist-piNTCF simulations (hist-piNTCF minus historical) over 1985-2014.

At the seasonal scale, biophysical temperature feedback is more complicated (Fig. 2g–i). In boreal regions, feedback differs between the cold and warm seasons. In the cold season, because baseline dT/dLAI is positive, LAI enhancement causes warming. However, the concurrent decline in dT/dLAI causes cooling, and this cooling surpasses warming from LAI responses, yielding net cooling (up to −0.05 °C). In contrast, in the warm season, baseline dT/dLAI is negative; therefore, LAI enhancement causes cooling. However, the concurrent increase in dT/dLAI causes warming, and this warming surpasses cooling from LAI responses, yielding net warming (up to 0.01 °C). Unlike boreal regions, temperate and tropical regions display a unimodal pattern of seasonal feedback. Specifically, in the warm season, cooling from dT/dLAI responses is partly offset by warming from LAI responses, resulting in net cooling. In the cold season, feedback from both LAI and dT/dLAI responses is relatively small.

Comparison of vegetation biophysical and biochemical feedback

Previous studies have demonstrated that aerosol decline can influence vegetation CO2 uptake2,3,4, which may further influence the global climate via changes in the atmospheric CO2 concentration. This raises a critical question regarding the relative importance of biochemical versus biophysical feedback from vegetation responses to aerosol decline in terms of climate mitigation. To answer this question, we further examine impacts of aerosol decline on terrestrial gross primary productivity (GPP), net primary productivity (NPP) and net ecosystem productivity (NEP).

GPP, an indicator of atmospheric CO2 removal through photosynthesis, exhibits contrasting latitudinal responses to aerosol decline (Fig. 4a). Boreal and temperate regions show widespread GPP increases, whereas tropical regions show decreases, a pattern that is consistent with LAI responses (Fig. 1b). NPP, which is calculated as GPP minus plant autotrophic respiration, responds to aerosol decline similarly to GPP but with attenuated magnitude (Fig. 4b). These patterns of GPP and NPP responses align with those reported in previous modeling studies8,10. NEP, which is calculated as NPP minus soil heterotrophic respiration, shows a much weaker response across most regions (Fig. 4c). The NPP and NEP responses are weaker than the GPP response because aerosol decline-induced warming enhances plant and soil respiration, offsetting the increased CO₂ uptake by vegetation. On average, responses of GPP, NPP and NEP to the idealized aerosol emission reductions are 0.071 ± 0.086, 0.008 ± 0.045 and −0.006 ± 0.018 gCO2·m-2·day−1, respectively, in vegetated areas.

Fig. 4: Impacts of idealized aerosol emission reductions on terrestrial carbon fluxes.
Fig. 4: Impacts of idealized aerosol emission reductions on terrestrial carbon fluxes.
Full size image

a–c Impacts of aerosol decline on terrestrial gross primary productivity (GPP; unit: gCO2·m−2·day−1; a), net primary productivity (NPP; unit: gCO2·m−2·day-1; b) and net ecosystem productivity (NEP; unit: gCO2·m−2·day−1; c). d The equivalent temperature effect from the NEP response, i.e., biochemical temperature feedback. e Relative strength of biochemical (BCH) and biophysical (BPH) feedback from vegetation responses to aerosol decline. f Comparison of regional mean BPH (blue bar) and BCH (orange bar) in boreal, temperate, and tropical vegetated areas, with bars and error bars representing the multimodel mean and intermodel standard deviation, respectively. In (a-d), shading denotes the multimodel mean, and black dots denote where at least four out of the five models agree on the sign of the multimodel mean. In (a–c), imbedded bar plots show regional averages across boreal (blue), temperate (gray), and tropical (orange) vegetated areas, with bars and error bars representing the multimodel mean and intermodel standard deviation, respectively. The results are derived from the comparison between the historical and hist-piNTCF simulations (hist-piNTCF minus historical) over 1985-2014.

To facilitate comparisons between vegetation biochemical and biophysical feedback, we convert the NEP response to an equivalent temperature metric on the basis of grid-scale transient temperature sensitivity to global cumulative CO2 emissions (TCRE; see Methods). The negative global mean NEP response signifies diminished terrestrial carbon sinks following aerosol decline, leading to widespread warming (Fig. 4d). Compared to biophysical feedback, biochemical feedback is stronger in Central Africa and Southeast Asia, yet weaker in the Amazon and most boreal and temperate regions (Fig. 4e). On average, this biochemical warming is 0.005 ± 0.015 °C for boreal vegetated areas, 0.003 ± 0.010 °C for temperate vegetated areas and 0.003 ± 0.009 °C for tropical vegetated areas (Fig. 4f). Notably, at the local scale, while the models agree well on the sign of NEP response, they exhibit substantial spreads in magnitude. Consequently, their global mean values show considerable intermodel variability, leading to significant uncertainty in the derived temperature effect.

Biophysical feedback from vegetation responses to future aerosol pollution control

The above analysis reveals vegetation biophysical feedback resulting from reducing current anthropogenic aerosol emissions to preindustrial levels—an idealized scenario unlikely to occur in reality. Next, we investigate vegetation biophysical feedback under a future scenario in which anthropogenic aerosol emissions are reduced in a more feasible way. For this purpose, we use two additional coupled simulations: the ssp370 simulation from the Scenario Model Intercomparison Project40 (ScenarioMIP) and the ssp370-lowNTCF simulation from the AerChemMIP37 (see Methods). The ssp370 simulation spans 2015–2100 under a medium-to-high forcing pathway with weak controls on aerosol and ozone emissions40 (Supplementary Figs. S1 and S6). The ssp370-lowNTCF simulation spans 2015–2055 under a pathway identical to that of the ssp370 simulation but with stringent controls on aerosol and ozone emissions37 (Supplementary Figs. S1 and S6). Differences between the two simulations (ssp370-lowNTCF minus ssp370) over 2026–2055 isolate the effect of future emission reductions. Here, considering the availability of required output variables, only four models (BCC-ESM1, CNRM-ESM2-1, NorESM2-LM, and UKESM1-0-LL) are available and averaged to produce the multimodel mean for analysis. The results are considered robust when at least three of the four models agree on the sign of the multimodel mean, indicated by black dots in the figures.

If stringent air pollution control is implemented in the future, aerosol loading is projected to decrease over continents (Fig. 5a), although the strength is weaker than that observed in the idealized scenario. Nevertheless, considerable reductions in aerosol optical depth (up to -0.2) can be found in some regions, such as eastern China and India. The aerosol decline results in robust changes in annual LAI across vegetated areas (Fig. 5b), with LAI increasing in boreal (0.03 ± 0.03 m2·m−2) and temperate (0.02 ± 0.02 m2·m−2) regions but decreasing in tropical regions (-0.01 ± 0.03 m2·m−2). Furthermore, aerosol decline also reduces annual dT/dLAI in most vegetated areas, except in South Asia, Columbia, and western Australia (Fig. 5c). On average, dT/dLAI reductions are strongest in boreal regions (−0.0019 ± 0.0021 °C·m-2·m2), followed by temperate (-0.0013 ± 0.0026 °C·m−2·m2) and tropical (-0.0005 ± 0.0026 °C·m−2·m2) regions.

Fig. 5: Annual effects of future aerosol emission reductions.
Fig. 5: Annual effects of future aerosol emission reductions.
Full size image

a–c, e Impacts of aerosol decline on the aerosol optical depth at 550 nm (unitless; a), leaf area index (LAI; unit: m2·m-2; b), the temperature sensitivity to LAI (dT/dLAI; unit: °C·m-2·m2; c), and terrestrial net ecosystem productivity (NEP; unit: gCO2·m-2·day−1; e) d, Biophysical feedback on the near-surface air temperature. f The equivalent temperature effect from the NEP response, i.e., biochemical temperature feedback. In each panel, shading denotes the multimodel mean, and black dots denote where at least three out of the four models agree on the sign of the multimodel mean. In (b–e), imbedded bar plots show regional averages across boreal (blue), temperate (gray), and tropical (red) vegetated areas, with bars and error bars representing the multimodel mean and intermodel standard deviation, respectively. The results are derived from the comparison between the ssp70 and ssp370-lowNTCF simulations (ssp370-lowNTCF minus ssp370) over 2026-2055.

As a result of these changes in LAI and dT/dLAI, vegetation creates complex biophysical feedback on the local near-surface air temperature (Fig. 5d). Cooling (up to <−0.03 °C) is dominant in eastern China, Eurasia, North America, Central Africa and southeastern South America, whereas warming (up to 0.02 °C) prevails in Southeast Asia, Columbia and Australia. In general, the sign of feedback largely aligns with dT/dLAI responses, i.e., warming where dT/dLAI increases and cooling where it decreases, implying the dominant role of dT/dLAI responses in determining feedback. Meanwhile, stringent air pollution control is projected to reduce global NPP by 0.002 ± 0.015 gCO2·m−2·day−1, despite increases in a few regions such as eastern China (Fig. 5e). This diminished NPP at the global scale further causes a slight warming of 0.001 to 0.005 °C over most continental regions (Fig. 5f). In summary, impacts of future air pollution control on LAI and dT/dLAI and the resulting biophysical and biochemical temperature feedback are largely consistent with those observed in the idealized scenario shown in the previous sections but with weaker strengths.

Discussion

In this study, we investigate biophysical climate feedback from vegetation responses to aerosol decline, which has been largely overlooked in previous studies. At the annual scale, this feedback leads to local cooling in boreal and temperate regions but local warming in tropical regions (excluding the Amazon). This latitudinal asymmetric pattern contrasts with that of biophysical feedback from vegetation responses to rising atmospheric CO2 concentrations, which has been reported to cause warming globally22,23,24,27. Aerosol decline causes warming (Supplementary Fig. S3c), but in boreal and temperate vegetated areas, the resulting biophysical cooling from vegetation (Fig. 3a) can offset up to ~7% of this warming effect at the local scale. However, in certain tropical regions (e.g., Central Africa) or seasons (e.g., boreal summers), vegetation feedback produces a warming effect that may exacerbate aerosol-decline induced warming. As such, from the perspective of counteracting aerosol-related radiative warming, the role of vegetation biophysical feedback seems limited and, in some cases, counterproductive. These findings underscore the need for additional mitigation strategies to address the warming resulting from aerosol decline due to air pollution control.

Although vegetation biophysical feedback can offset only a small portion of the warming from aerosol decline, it nonetheless remains a valuable component of climate mitigation. First, in certain regions (e.g., China), biophysical feedback can be considerable and even stronger than biochemical feedback emphasized in recent studies31,32 in terms of climate mitigation. While the well-mixed nature of CO2 means that biochemical feedback operates globally, biophysical feedback remains significant at the local and regional scales. Second, our reported biophysical feedback from vegetation responses to aerosol decline is comparable in magnitude to that caused by other drivers. For example, satellite observations have revealed widespread vegetation greening in recent decades due to rising CO2 levels, climate shifts, land use and management, and nitrogen deposition15,41,42. Local biophysical feedback from this greening trend is estimated to have a regional mean value of −0.003 °C for boreal regions, −0.027 °C for temperate regions and −0.022 °C for tropical regions16. In comparison, our estimated local biophysical feedback from vegetation responses to complete removal of anthropogenic aerosol emissions is −0.008 °C, −0.004 °C and 0.001 °C, respectively, in the corresponding regions. Notably, previous estimates mostly rely on the satellite-based land surface temperature16,39, whereas our study uses the near-surface air temperature. Since vegetation feedback on the latter is merely 50% or less than that on the former43,44,45, our reported vegetation feedback may be more comparable in strength to that from other drivers if the same temperature metric are used.

Our earlier study reported enhanced cooling effects from temperate afforestation under low-aerosol emission scenarios46, although the roles of LAI and dT/dLAI responses remained unquantified. In this study, we elucidate the regionally divergent mechanisms driving biophysical feedback from vegetation responses to aerosol decline. In general, in boreal and temperate regions, annual cooling is due to the combined effects of increasing LAI and decreasing dT/dLAI. In contrast, in most tropical regions, annual warming is due to decreasing LAI, which is partially counteracted by cooling from decreasing dT/dLAI. The dT/dLAI shift can be further attributed to shifts in the albedo-driven effect and the evapotranspiration-driven effect. In boreal regions, the dT/dLAI decline (i.e., less positive) mainly results from diminished snow cover following aerosol decline, which alleviates the albedo-driven warming effect in the cold season. In temperate and tropical regions, the dT/dLAI decline (i.e., more negative) is attributed to the enhanced evapotranspiration-driven cooling effect, driven by higher solar radiation, warmer temperature and larger leaf area following aerosol decline. These findings highlight complex responses of LAI and dT/dLAI to aerosol decline and the resulting spatial heterogeneity in temperature feedback. More importantly, this study demonstrates that vegetation can yield biophysical feedback on local temperature even in the absence of direct human interventions (e.g., afforestation) under low-aerosol emission scenarios.

As noted in the previous sections, in both the idealized and future scenarios, not only aerosols but also ozone emissions are reduced. This raises the following question: Do the vegetation response and feedback result from aerosol decline, ozone decline, or both? To answer this question, we replace the hist-piNTCF simulation with the hist-piAer simulation from the AerChemMIP37, which only maintains aerosol emissions at preindustrial levels while keeping other model configurations identical (see Methods). We find that excluding the ozone effect has almost no influence on the main results (Supplementary Figs. S7–S9), indicating that the vegetation response and feedback primarily stem from aerosol decline rather than ozone decline. However, we emphasize that this result does not negate the known impact of ozone on vegetation. Previous studies have demonstrated that ozone can enter leaves via stomata and damage photosynthesis by oxidizing plant cells, thereby suppressing vegetation growth and transpiration13,47,48. Our simulations do not capture these ozone effects, largely because most CMIP6 models lack a complete representation of ozone vegetation damage processes. Thus, future studies using models with fully coupled ozone‒vegetation interactions47,49,50 would be valuable for assessing ozone’s impact on vegetation and its potential climate feedbacks, which is beyond the scope of this study.

We acknowledge several limitations of this study that should be mentioned. First, we use an empirical model to estimate dT/dLAI. While the model captures the effects related to changes in albedo and evapotranspiration—two important biophysical pathways by which vegetation changes influence climate15,16,17,18,39, they represent only local processes51,52,53. A growing number of studies have revealed that vegetation changes can exert far-reaching impacts on regional or even remote climates via changes in cloud formation54,55,56 and large-scale atmospheric circulations57,58,59. However, this study does not account for these nonlocal processes and their associated climate consequences. Thus, biophysical climate feedback from vegetation responses to aerosol decline presented here might be conservative and should be interpreted as local effects.

Second, we do not consider the noncarbon cycle biochemical effect of vegetation. For example, vegetation emits biogenic volatile organic compounds (BVOCs), which can be oxidized to form secondary organic aerosols (SOAs)60. BVOC emission rates are also climate sensitive, with warmer conditions generally promoting higher emissions61,62,63. Higher LAI resulting from reduced aerosol levels, combined with the warming induced by aerosol decline, tends to promote SOA formation, leading to cooling both directly by reducing incoming solar radiation and indirectly by acting as cloud condensation nuclei that facilitate cloud droplet formation46,64,65. Furthermore, these BVOC and SOA responses to LAI expansion can, in turn, modulate vegetation biophysical feedback through cloud and radiation feedback66. Therefore, a comprehensive evaluation of biochemical climate feedback resulting from vegetation responses to aerosol decline requires a fully coupled model that represents vegetation’s noncarbon biochemical processes and their interactions with climate.

Third, given the model output required to calculate vegetation biophysical feedback, we use only a limited number of CMIP6 models for analysis. This implies that our results may be biased towards the models used. Additionally, prior evaluations have shown considerable intermodel spread in simulated LAI variations during both historical and future periods67,68,69, as well as in vegetation biophysical climate feedback70,71,72, across CMIP6 models. As such, more models, along with the application of observation-based constraint methods, are needed to provide a more reliable simulation of biophysical climate feedback resulting from vegetation responses to aerosol changes.

Finally, our analysis focuses solely on the period when differences in aerosol emissions between paired simulations reach their maximum. During this period, the identified climate changes primarily reflect the rapid response to aerosol variations, driven by immediate radiative and atmospheric adjustments. However, previous simulation studies have shown that the climate system may also respond to aerosol changes over longer timescales, ranging from years to centuries73,74,75. This slower climate response is mainly linked to the gradual evolution of ocean heat content and ice sheets in response to aerosol forcing. Additionally, vegetation exhibits inertia and may require extended periods to fully adjust to changes in radiation and climate76,77,78. Given the delayed climate responses to aerosols and their subsequent legacy effects on vegetation, long-term simulations are necessary to fully capture biophysical feedback arising from vegetation responses to aerosol changes.

Our findings have implications for designing synergistic air quality control and climate mitigation strategies. Vegetation-based climate solutions, such as afforestation and reforestation, have long been regarded as cost-effective approaches for mitigating climate change through biophysical and biochemical pathways. This study underscores that the climate mitigation potential of vegetation is not static but is instead regulated by air quality levels. In extratropical regions, aerosol decline can strengthen the climate mitigation potential of vegetation, primarily via biophysical pathways. In contrast, in tropical regions, aerosol decline may impair the climate mitigation potential of vegetation through both biophysical and biochemical pathways. This regional disparity points to an emerging environmental inequity: tropical nations, already disproportionately vulnerable to climate change, could face additional climate mitigation burdens in a low-aerosol future. Thus, tropical nations require nuanced, location-specific policies that carefully balance air quality improvements with climate mitigation goals.

Methods

Numerical simulations

To isolate the aerosol effect, we use five fully coupled simulations with varying anthropogenic emission scenarios, including the historical simulation from the Coupled Model Intercomparison Project Phase 636 (CMIP6), the hist-piNTCF, hist-piAer and ssp370-lowNTCF simulations from the Aerosol Chemistry Model Intercomparison Project37 (AerChemMIP), and the ssp370 simulation from the Scenario Model Intercomparison Project40 (ScenarioMIP).

The historical simulation serves as the baseline simulation covering the period of 1850–2014. This simulation is driven by evolving and externally imposed natural (e.g., solar variability and volcanic aerosols) and anthropogenic (e.g., greenhouse gas emissions, aerosol emissions, and land use changes) forcings that are largely consistent with observations. The emissions of anthropogenic aerosols and reactive species are prescribed according to the datasets generated by the Community Emissions Data System79. The hist-piNTCF simulation is identical to the historical simulation, except that anthropogenic aerosol and ozone precursor emissions are fixed at preindustrial (1850) levels. Thus, differences between the historical and hist-piNTCF simulations (i.e., hist-piNTCF minus historical) isolate the effect of complete removal of anthropogenic emissions. We focus on the last 30 years (i.e., 1985–2014) because the emission difference reaches a maximum during this period (Supplementary Figs. S1 and S2). Five models (i.e., BCC-ESM1, CNRM-ESM2-1, NorESM2-LM, TaiESM1, and UKESM1-0-LL) provide the required output variables for both simulations (Supplementary Table S1) and are used for analysis.

Notably, in addition to aerosol emissions, the historical and hist-piNTCF simulations also differ in terms of ozone emissions. To eliminate the potential ozone effect, we use an additional simulation hist-piAer, which is identical to the historical simulation except that only anthropogenic aerosol precursor emissions are maintained at preindustrial levels. Differences between the historical and hist-piAer simulations (i.e., hist-piAer minus historical) thus isolate the effect attributable solely to aerosol decline. For this comparison, five models (BCC-ESM1, IPSL-CM6A-LR, MPI-ESM-1-2-HAM, NorESM2-LM, and UKESM1-0-LL) provide the necessary outputs and are used for analysis. Our analysis demonstrates that the ozone effect is negligible, indicating that the vegetation response and feedback mainly result from aerosol decline. Therefore, we present the results derived from the historical and hist-piNTCF comparison in the main text (Figs. 1–4) and those from the historical and hist-piAer comparison in the supplementary material (Supplementary Figs. S7–S9).

For future scenarios (2015–2100), the ssp370 simulation serves as the baseline simulation, which follows the Shared Socioeconomic Pathway 3 (regional rivalry) along with the Representative Concentration Pathway 7.0. This scenario represents a medium-to-high forcing pathway with substantial greenhouse gas emissions and land use changes (particularly forest cover loss). Short-lived climate pollutants (e.g., aerosols and ozone) remain at the highest levels due to weak air quality controls. The ssp370-lowNTCF simulation is identical to the ssp370 simulation but assumes strong air quality policies, leading to substantial reductions in anthropogenic aerosol and ozone precursor emissions. To reduce computational costs, the ssp370-lowNTCF simulation runs only until 2055, when air quality policy divergence is expected to peak. Differences between the ssp370 and ssp370-lowNTCF simulations (i.e., ssp370-lowNTCF minus ssp370) over the period of 2026–2055 thus isolate the effect of future aerosol and ozone decline under stringent air quality measures (Supplementary Figs. S1 and S6). For this comparison, four models (BCC-ESM1, CNRM-ESM2-1, NorESM2-LM, and UKESM1-0-LL) provide the necessary outputs and are used for future projection.

For each model, if only one realization is available for the baseline and experimental runs (e.g., historical versus hist-piNTCF), that single member is used; otherwise, the first member is selected to facilitate intermodel comparison (Supplementary Table S2). All model outputs are analysed at their native spatial resolutions before being interpolated to a common resolution of 3° × 3° to compute the multimodel mean. The results at each grid cell are considered robust if at least 75% of the models (i.e., more than four out of five models for the historical scenario, or more than three out of four models for the future scenario) agree on the sign of the multimodel mean, as indicated by black dots in the figures.

Estimation of the temperature sensitivity to leaf area changes

To quantify the biophysical temperature feedback from vegetation responses to aerosol decline, the temperature sensitivity to leaf area index (LAI) changes (dT/dLAI; unit: °C·m−2·m2) is needed. dT/dLAI quantifies the local near-surface air temperature change induced by each unit increase in LAI. We calculate dT/dLAI in vegetated areas (LAI ≥ 0.1 m2·m−2) using an empirical model proposed by Alkama et al. 38. This model is built on satellite-observed dT/dLAI data that are retrieved through the space and time analogy method80. In this method, dT/dLAI is quantified as the temperature trend difference between pixels undergoing LAI changes and nearby pixels without LAI changes80.

This empirical model is superior in its minimal input requirements while effectively accounting for the impact of vegetation changes on local temperature through biophysical processes (e.g., albedo and evapotranspiration). Specifically, under snow-covered (snow cover fraction ≥1%) conditions, dT/dLAI is dominated by the effect arising from albedo changes, which are further amplified by snow cover. In this case, dT/dLAI is formulated as a function of monthly snow cover (snc; unit: %) and downwards solar radiation (rsds; unit: W·m−2), expressed as follows:

$$\begin{array}{lll}\frac{dT}{dLAI}=-1.66\times {10}^{-5}\times sn{c}^{2}-2.14\times {10}^{-7}\times rsd{s}^{2}+1.56\times {10}^{-5}\\\quad\quad\quad\,\,\,\,\times snc\times rsds+2.90\times {10}^{-3}\times snc-8.23\times {10}^{-4}\times rsds-0.011\end{array}$$
(1)

Under snow-free (snow cover fraction <1%) conditions, dT/dLAI is dominated by changes in the partitioning of surface available energy into latent and sensible heat fluxes. In this case, dT/dLAI is formulated as a function of monthly evapotranspiration (et; unit: mm·day−1), which can be calculated on the basis of the model output latent heat flux (hfls; unit: W·m-2) and the heat of vaporization (2.5×106 J·kg-1), along with downwards solar radiation (rsds; unit: W·m-2), expressed as follows:

$$\begin{array}{l}\frac{dT}{dln(LAI)}=3.64\times {10}^{-3}\times e{t}^{2}+1.17\times {10}^{-5}\times rsd{s}^{2}-5.80\times {10}^{-5}\\\qquad\quad\,\,\,\,\times et\times rsds+8.51\times {10}^{-2}\times et-6.74\times {10}^{-3}\times rsds+0.42\end{array}$$
(2)

The use of ln(LAI) in the denominator of the left-hand term of Eq. (2) accounts for the sensitivity of dT/dLAI to LAI (i.e., the dT/dLAI strength diminishes as LAI increases), as observed in both satellite observations16,38,39 and simulations18. Furthermore, the use of evapotranspiration in Eq. (2) implies that the vegetation physiological response (e.g., stomatal conductance) to aerosol decline is implicitly considered. dT/dLAI is first calculated at the monthly scale and then aggregated into th e annual value.

It is worth noting that, apart from the albedo and evapotranspiration pathways, vegetation changes can also affect local temperatures through modifications in aerodynamic roughness, which in turn alter surface turbulent fluxes60. However, the impact of roughness is often secondary and varies spatially. It tends to be more significant in boreal regions during winter, when the albedo mechanism plays a dominant role, but is relatively smaller in tropical and temperate ecosystems81. Moreover, isolating the roughness effect using only satellite data is difficult due to the collinearity between roughness changes and other vegetation characteristics. Therefore, the model includes only the albedo and evapotranspiration effects, which account for the dominant portion of vegetation-induced surface temperature variability at regional and global scales16,45,82.

The performance of the empirical model has been validated against satellite observations, and the model outcome can reasonably reproduce the satellite-observed dT/dLAI in terms of both spatial patterns and seasonal variations38. The model outcome is also broadly consistent with the satellite-observed biophysical impacts of LAI expansion16,39 or forest cover gains80,82,83,84 on land surface temperature. Notably, since the model estimates vegetation feedback on near-surface air temperature (i.e., 2-m temperature), its strength is generally weaker than that of satellite-based estimates that focus on feedback on land surface temperature16,39.

Estimation of biophysical feedback from vegetation responses to aerosol decline

Vegetation biophysical feedback on local near-surface air temperature (ΔT; unit: °C) depends on both LAI and dT/dLAI responses to aerosol decline, expressed as follows16,38,39:

$$\Delta T=\Delta \left(\frac{dT}{dLAI}\times LAI\right)$$
(3)

where Δ denotes the response to aerosol decline, which is calculated as the difference between simulations (e.g., historical and hist-piNTCF).

To further quantify the individual contributions from LAI and dT/dLAI responses to total feedback, we decompose the right-hand side of Eq. (3) into three components, expressed as follows:

$$\Delta T=\left(\frac{{dT}}{{dLAI}}\right)\times \Delta {LAI}+{LAI}\times \Delta \left(\frac{{dT}}{{dLAI}}\right)+\varepsilon$$
(4)

where \(\left(\frac{{dT}}{{dLAI}}\right)\times \Delta {LAI}\) denotes feedback from LAI responses, \({LAI}\times \Delta (\frac{{dT}}{{dLAI}})\) denotes feedback from dT/dLAI responses, and ε denotes the residual term. The individual contribution ratios of \(\left(\frac{{dT}}{{dLAI}}\right)\times \Delta {LAI}\) (ρLAI; unit: %), \({LAI}\times \Delta (\frac{{dT}}{{dLAI}})\) (ρdT/dLAI; unit: %) and ε (ρε; unit: %) to total feedback can be expressed as follows:

$${\rho }_{{LAI}}=\frac{\left|\left(\frac{{dT}}{{dLAI}}\right)\times \Delta {LAI}\right|}{\left|\left(\frac{{dT}}{{dLAI}}\right)\times \Delta {LAI}\right|+\left|{LAI}\times \Delta (\frac{{dT}}{{dLAI}})\right|+\left|\varepsilon \right|}\times 100 \%$$
(5)
$${\rho }_{{dT}/{dLAI}}=\frac{\left|{LAI}\times \Delta (\frac{{dT}}{{dLAI}})\right|}{\left|\left(\frac{{dT}}{{dLAI}}\right)\times \Delta {LAI}\right|+\left|{LAI}\times \Delta (\frac{{dT}}{{dLAI}})\right|+\left|\varepsilon \right|}\times 100 \%$$
(6)
$${\rho }_{\varepsilon }=\frac{\left|\varepsilon \right|}{\left|\left(\frac{{dT}}{{dLAI}}\right)\times \Delta {LAI}\right|+\left|{LAI}\times \Delta (\frac{{dT}}{{dLAI}})\right|+\left|\varepsilon \right|}\times 100 \%$$
(7)

If \({\rho }_{{LAI}} > {\rho }_{{dT}/{dLAI}}\), we consider that LAI responses dominate total feedback; otherwise, dT/dLAI responses dominate total feedback. ρε is mostly below 10% (Supplementary Fig. S10), demonstrating that Eq. (4) effectively attributes vegetation biophysical feedback to LAI and dT/dLAI responses.

Converting biochemical feedback to the equivalent temperature effect

In addition to biophysical feedback, vegetation can produce biochemical feedback on climate via changes in CO2 uptake following aerosol decline. To facilitate the comparison between biochemical and biophysical feedback, we convert biochemical feedback into an equivalent temperature metric \(\Delta {T}^{{BCH}}\). This metric quantifies the temperature effect resulting from changes in terrestrial carbon fluxes following aerosol decline. \(\Delta {T}^{{BCH}}\) (unit: °C) in a grid cell (i, j) can be calculated as follows35,46:

$$\Delta {T}^{{BCH}}(i,j)=-\triangle {NEP}\times {TCRE}(i,j)$$
(8)

where ΔNEP denotes the accumulated net ecosystem productivity (NEP) change (unit: Gt CO2·m-2) over global land owing to aerosol decline during the analysis period (1985–2014), and TCRE(i, j) denotes the transient temperature sensitivity in the grid cell (i, j) to global cumulative CO2 emissions (unit: °C per GtCO2). The negative sign before ΔNEP accounts for the cooling associated with increased NEP (i.e., positive ΔNEP) and the warming associated with decreased NEP (i.e., negative ΔNEP).

TCRE can be derived from the 1pctCO2 simulation of CMIP636, which is a coupled simulation spanning a minimum length of 150 years. In this simulation, the atmospheric CO2 concentration increases by 1% per year from preindustrial levels, whereas other external forcings remain constant. The TCRE in a grid cell (i, j) is then calculated as follows35,46:

$${\rm{TCRE}}(i,j)=\frac{({{\rm{T}}}_{2{CO}2}\left(i,j\right)-{{\rm{T}}}_{0}(i,j))}{2.27\times ({C}_{2{CO}2}-{C}_{0})\times 7.81}$$
(9)

where T2CO2(i, j) is the 20-year mean near-surface air temperature (unit: °C) in the grid cell (i, j) centered on the time when the atmospheric CO2 concentration doubles, T0(i, j) is the mean near-surface air temperature (unit: °C) in the grid cell (i, j) in the first 20 years, corresponding to the equilibrium value at preindustrial CO2 levels, C0 is the initial global atmospheric CO2 concentration (286 ppm), and C2CO2 is the doubling of C0 (572 ppm). The factor of 2.27 in the numerator is the reciprocal of the airborne fraction of CO2 (0.44), which is defined as the fraction of anthropogenic CO2 emissions that remain in the atmosphere1. The factor of 7.81 in the numerator converts ppm to GtCO2 in units. The global mean TCRE across the models used in this study ranges from 1.4 to 2.1 °C per 1000 PgC, which falls well within the likely range of 1.0 to 2.3 °C per 1000 PgC1.

Climate classification

We classify boreal, temperate, and tropical regions on the basis of the updated global Köppen-Geiger climate classification map85 for the present-day period (1980–2016). This map integrates an ensemble of high-resolution climate datasets, including three air temperature sources (WorldClim V1 and V2 and CHELSA V1.2) and four precipitation sources (WorldClim V1 and V2, CHELSA V1.2, and CHPclim V1). The classification comprises five major climate zones: tropical, arid, temperate, boreal, and polar. However, we focus exclusively on tropical, temperate, and boreal regions, as vegetated areas (LAI ≥ 0.1 m²·m⁻²) are scarce in arid and polar zones (Fig. 1f).