The EMAC (ECHAM/MESSy Atmospheric Chemistry) model is a numerical chemistry–climate modelling system that includes submodels that represent tropospheric and middle-atmospheric processes, as well as their interactions with oceans, land, and human activities. It originally combined ECHAM, an atmospheric general circulation model (GCM) , with the Modular Earth Submodel System (MESSy) framework and philosophy, modularising physical processes and most of the infrastructure into submodels that can be further developed to improve existing process representations. New submodels can also be added to represent new or alternative process representations.

Aerosols are treated using the submodel GMXe , in which aerosol microphysics are characterised by seven interactive log-normal modes that span the typical size range of aerosol species. These modes are further categorised into four hydrophilic aerosol modes (nucleation, Aitken, accumulation, and coarse) and three hydrophobic aerosol modes (Aitken, accumulation, and coarse). The representation of all aerosols assumes that particles are spherical. The properties of aerosols in each mode are fully determined by the total mass (the internal mixture of contributing species), density, number concentration, median radius, and width of the log-normal distribution. After each simulation step, aerosols may transfer between modes depending on size changes.

Organic aerosol species are additionally described by ORACLE, a submodel for the description of organic aerosol composition and evolution , taking into account the partitioning between aerosols and the gas phase based on the volatility basis set (VBS) framework . ORACLE describes the following organic aerosols (OAs): secondary organic aerosols from the oxidation of anthropogenic VOCs (aSOAs) and biogenic VOCs (bSOAs); primary organic aerosols from emissions arising from fossil fuels (fPOAs) and biofuel combustion and biomass burning (bbPOAs); and secondary organic aerosols arising from the subsequent photochemical oxidation of fPOAs and bbPOAs (fSOAs and bbSOAs, respectively) . ORACLE treats mass yields of secondary organic aerosol (SOA) with respect to different lumped VOCs at varying levels of saturation concentration (C*), expressed in µg m⁻³, at 298 K, based on an assumed particle density of 1.5 g cm⁻³. Isoprene exhibits mass yields across the range of C*, with a peak of 0.03 at 10 µg m⁻³, before declining sharply to 0.015 at 100 µg m⁻³ and reaching zero at 1000 µg m⁻³. Monoterpenes show significantly higher yields, starting at 0.107 for a low value of C* (1 µg m⁻³) and peaking at 0.600 for a higher value of C* (1000 µg m⁻³), highlighting the much greater contribution of monoterpenes to SOA formation, especially under conditions of higher saturation concentration. More details on SOA mass yields in ORACLE can be found in .

ORACLE employs a simple photochemical-ageing scheme that effectively models the combined impacts of the fragmentation and functionalisation of organic compounds. The module not only predicts the mass concentration of organic aerosol (OA) components but also predicts their oxidation state (expressed as O : C), enabling their categorisation into primary OAs (POAs – chemically unprocessed); freshly formed secondary OAs (SOAs), exhibiting a low oxygen content; and aged SOAs (highly oxygenated). By explicitly simulating the chemical conversion of OA from initial emissions to a highly oxygenated state during photochemical ageing, ORACLE facilitates the tracking of changes in OA hygroscopicity resulting from these reactions. This enables the computation of the OA particle's capability to serve as cloud condensation nuclei. The output from the ORACLE model, based on the described setup, has been compared with observational data from tropical regions . We found that ORACLE provides surface OA concentrations within 60 % of the observed values recorded over these tropical forest regions. A thorough evaluation of ORACLE against aerosol mass spectrometer (AMS) measurements is provided in . Details on the implementation of ORACLE (v2.0) in EMAC, along with a comprehensive model evaluation, can be found in .

GMXe treats new-particle formation (NPF) based on temperature, relative humidity, and sulfuric acid (H₂SO₄) concentration . In this setup, organics do not contribute to NPF but rather participate in condensation through the VBS framework in ORACLE (as described above), where volatilities are governed by the oxidation of organic precursors.

Heterogeneous and gas-phase chemistry are treated through the MECCA submodel , employing version 1 of the Mainz Isoprene Mechanism (MIM1) as the chemical mechanism , which includes over 100 gas-phase species and more than 250 reactions. MIM1 includes the following BVOC oxidation pathways: isoprene + OH, isoprene + O₃, and monoterpene oxidation (lumped species) with OH, O₃, NO₃, and O(¹D). Dry-deposition, sedimentation, and wet-deposition processes are simulated using the submodels DDEP, SEDI both , and SCAV , respectively.

Convective cloud processes are taken into account based on the approach proposed by , utilising the convection schemes from and . Convective cloud microphysics is solely based on temperature and moisture profiles and does not account for the influence of aerosols on liquid droplet or ice formation processes. The vertical velocity distribution used for aerosol activation by grid-scale clouds in EMAC is calculated as the sum of the grid's mean vertical velocity and the turbulent contribution, as detailed by . Large-scale stratiform clouds are described by the CLOUD submodel, which, in the applied configuration, incorporates a two-moment cloud microphysics scheme for cloud droplets and ice crystals, as detailed by , , and . CLOUD solves prognostic equations for specific humidity, the liquid cloud mixing ratio, the ice cloud mixing ratio, cloud droplet number concentration (CDNC), and ice crystal number concentration (ICNC).

CLOUD incorporates a prognostic cloud droplet nucleation process to represent aerosol–cloud interactions in large-scale clouds (excluding warm convective clouds) using aerosol information provided by GMXe. This prognostic nucleation scheme is based on an aerosol activation parameterisation called the “unified dust activation framework” (UAF). The UAF is based on the cloud activation parameterisation from and includes dust–pollutant interactions during the activation process by taking into account the hydrophilicity of the dust using an adsorption theory, as well as the acquired hygroscopicity through Köhler theory. This routine provides values for the hygroscopicity parameter (κ) and the cloud condensation nuclei (CCN) number concentration at levels of 0.2 % and 0.4 % supersaturation for each aerosol size mode .

For the evaluation of the radiative effects from clouds and aerosols, we employ methods from . The calculation of the net radiative effect from aerosol–radiation interactions (RE_ari) involves determining the difference between the net top-of-the-atmosphere shortwave radiative flux (F) and the radiative flux, excluding the scattering and absorption of solar radiation caused by the aerosols (Fclean). Fclean is computed in a separate radiation call within the radiation submodel RAD. Similarly, the radiative effect resulting from aerosol–cloud interactions (RE_aci) is derived by assessing the difference between Fclean and the flux that disregards the scattering and absorption caused by both clouds and aerosols (Fclear-sky, clean).

EMAC includes land surface and vegetation models, enabling comprehensive studies of vegetation–atmosphere interactions. The Lund–Potsdam–Jena General Ecosystem Simulator (LPJ-GUESS) was the first vegetation model integrated into EMAC , and more recently, the Jena Scheme for Biosphere–Atmosphere Coupling in Hamburg (JSBACH) has also been incorporated as a submodel , further expanding EMAC’s capacity to simulate land surface processes. In this work, we rely exclusively on LPJ-GUESS for vegetation calculations.

In this work, we use the standard coupled configuration EMAC–LPJ-GUESS, in which the vegetation in LPJ-GUESS is entirely determined by the EMAC atmospheric state, soil type, N deposition, and CO₂ fluxes ; however, there is no feedback from the vegetation that influences the climate variables, except with respect to terrestrial BVOC emissions. The roughness length and albedo are kept as constant background values. The albedo is derived from satellite climatologies, while the roughness length is based on subgrid-scale orography and satellite-derived vegetation climatology. Vegetation changes do not feed back to the hydrological cycle. We use the native bucket model from ECHAM5, which employs fixed climatological vegetation . In this setup, BVOCs are interactive tracers that can be oxidised to form secondary organic aerosols. This means that BVOCs can influence the oxidant chemistry of the atmosphere; however, we do not quantify such impacts in this work and instead focus solely on aerosol changes.

After each simulation day, EMAC computes average daily values for 2 m temperature, net downward shortwave radiation, and total precipitation and passes these state variables to LPJ-GUESS. Vegetation information (i.e. the leaf area index, foliar density, leaf area density distribution, and PFT fractional coverage) from LPJ-GUESS is then fed back into EMAC for the calculation of BVOC emission fluxes using EMAC's BVOC submodels. In this study, the BVOC fluxes in EMAC are calculated using version 2.04 of the Model of Emissions of Gases and Aerosols from Nature (MEGAN) .

MEGAN is based on the work of , in which BVOC emission fluxes are calculated as a function of PFT-specific emission factors and non-dimensional activity factors. These activity factors consider sensitivities to the canopy environment, including parameters such as leaf area index (LAI), temperature, light, and leaf age. Notably, the current setup does not incorporate sensitivity to soil moisture. The “parameterised canopy environment emission activity” (PCEEA) algorithm is used, rather than an alternative detailed canopy environment model that calculates light and temperature values at each canopy depth. The PCEEA algorithm calculates the light sensitivity within the canopy as a function of the daily average above-canopy photosynthetic photon flux density (PPFD), the solar angle, and a non-dimensional factor describing PPFD transmission through the canopy.

This setup employs BVOC–aerosol–vegetation feedbacks, making vegetation and BVOC emissions sensitive to changes in temperature and above-canopy radiation (excluding diffused radiation) resulting from aerosol interactions. Nevertheless, these feedbacks are minimised by nudging meteorological fields towards observations. BVOC emissions from this model setup were evaluated and applied in other studies e.g..

The land cover scenarios were derived from version 3.2 of the History Database of the Global Environment (HYDE v3.2) . HYDE provides a wide range of land use products, encompassing both historical and projected data. To ensure an accurate representation, we relied on HYDE's cropland and grazing-land products from 2015, derived from high-resolution satellite data. These products were transformed into deforestation fraction maps (Fig. 1) to constrain the vegetation in the model. Three experiments were conducted to assess the impact of human-induced LCC on the natural land biosphere and atmospheric composition. The initial model run used simulated PNV without any deforestation. Additionally, two more model runs were conducted, incorporating deforestation. The first scenario aims to represent present-day land cover resulting from deforestation for cropland and grazing land. This is referred to as the scenario for deforested cropland and grazing land, i.e. the DCGL scenario (Fig. 1a). The DCGL scenario will sometimes be referred to as present-day deforestation or present-day land cover. The second scenario involves deforestation exclusively affecting cropland and is referred to as the deforested cropland (DCL) scenario (Fig. 1b). We use the DCL scenario to evaluate the potential impact of restoring all grazing land back to its natural state, essentially creating an extreme afforestation scenario while maintaining present-day croplands for agricultural food production.

All simulations were conducted over 12 years (2000–2011), with the initial 2 years excluded from the analysis to ensure proper spin-up and equilibrium states in the analysed data. The initial vegetation states for all simulations were taken from a previous non-chemistry 50-year run under similar atmospheric states. For this study, the simulations were performed at a T63L31 resolution, i.e. at a resolution of approximately 1.9° × 1.9° (or approx. 180×180 km at the Equator) with 31 vertical levels. The meteorological fields in the troposphere were nudged towards ERA-Interim reanalysis data for the years 2000–2012. Nudging the meteorology is important for preventing deviations in our simulations caused by feedbacks related to temperature and dynamics. By implementing nudging, these feedbacks are effectively suppressed, enabling us to evaluate changes in atmospheric states solely due to perturbed BVOC emissions resulting from land cover change.

The corresponding forcing at the sea surface (sea surface temperature (SST) and sea ice coverage (SIC)) is also inferred from the nudging data with continuous variation. Tracers were initialised using climatological data from previous simulations spanning the years from 2000 to 2020, while the CO₂ concentrations in the radiation and vegetation schemes were kept fixed at 384 ppmv, representing the year 2015.

Greenhouse gas mixing ratios, including N₂O, CH₄, CO₂, halons, and H₂, were prescribed at the surface level using data from the Chemistry-Climate Model Initiative (CCMI) for the year 2015 . Stratospheric H₂SO₄ mixing ratios were derived from a time series provided by the CCMI database. Biomass-burning emissions were simulated by the BIOBURN submodel, which imports dry-matter data from version 4.1 of the Global Fire Emissions Database (GFED4.1s) and employs emission factors from , also based on the year 2015. Anthropogenic emissions of black carbon (BC), carbon monoxide (CO), nitrogen oxides (NO_x), organic carbon (OC), sulfur dioxide (SO₂), alcohols, and organic gases were based on the Copernicus Atmosphere Monitoring Service (CAMS-GLOB-ANT v4.2 and CAMS-GLOB-AIR v1.1) . Degassing volcanic-climatology data were obtained from the Aerosol Comparisons between Observations and Models (AeroCom) project . Oceanic emissions and deposition were calculated online using the AIRSEA submodel for dimethyl sulfide (DMS), acetone (CH₃COCH₃), methanol (CH₃OH) (with an under-saturation of 6 %), and isoprene (C₅H₈). Natural emissions of ammonia (NH₃) were based on the Global Emissions Initiative (GEIA) database . Biogenic nitrogen oxide (NO), sea salt , and dust emissions were calculated online with the aid of the ONEMIS submodel , using corresponding climate states from EMAC. We note that all scenarios (PNV, DCGL, and DCL) use the same meteorology and emissions as described above. This means we do not account for past or future atmospheric forcing on vegetation and BVOC emissions (e.g. temperature changes). Instead, we focus only on the impact of perturbed BVOC emissions in a present climate state.

This section explores changes in the natural land biosphere caused by human deforestation with respect to crops and grazing land, based on HYDE land use data for 2015. We demonstrate how these alterations in the biosphere propagate to the atmosphere, affecting BVOC surface fluxes and impacting the aerosol burden and other atmospheric states.

In this section, we evaluate the changes resulting from the restoration of all grazing land to natural vegetation levels. In Sect. 3.1, the PNV scenario was used as the baseline to assess changes resulting from present-day land use cover (DCGL). In the following section, the DCGL scenario serves as the baseline, with the sensitivity run being the DCL scenario, including only deforested crop cultivation, as shown in Fig. 1b. These analyses therefore represent an extreme reforestation scenario.

Studies agree that land use changes predominantly affect BVOC emissions through the expansion of croplands in tropical regions, such as the Amazon, central Africa, and Southeast Asia . For example, show a reduction of approximately 24 % in isoprene emissions between 1901 and 2002. Additionally, reports that land cover changes from the 1850s to the 2000s resulted in a global decrease of approximately 35 % in BVOC emissions. In a comprehensive literature review exploring the influence of land use and land cover change (LULCC) on atmospheric composition and climate, it was estimated that from the preindustrial era to the present, LULCC has led to a decrease of 15 %–36 % in global isoprene emissions . report global decreases in isoprene and monoterpenes emissions of 87 % and 94 %, respectively, resulting from complete global deforestation. The 26 % decrease in BVOC emissions reported in this study is in line with findings reported in the literature, particularly because many studies analyse changes from preindustrial times, whereas our work focuses on changes resulting from human land use compared to natural vegetation. Although neither cropland nor grazing-land expansion was as significant as during or after the industrial era, it did occur on a smaller scale. Attention must also be given when comparing studies with and without temperature feedback, given the strong dependence of BVOC emissions on temperature. In this work, similar to the calculations in , temperature feedbacks were suppressed.

The global yearly isoprene and monoterpene emission budgets from this study are relatively high compared to global estimates observed in the literature, as shown in . Some atmospheric chemistry studies using MEGAN in EMAC, e.g. , employ a global scaling factor to dampen the global emissions to desired values. However, in this study, no scaling factors were applied, which means that the values reported here may be slightly overestimated.

The rapid oxidation of BVOCs yields oxygenated intermediate species near the surface, which act as precursors for the formation of bSOA. This explains the high abundance of bSOA in the lower atmosphere (see the vertical profiles in Fig. 4). As bSOA is dispersed in the atmosphere, concentrations typically decrease due to mixing and dilution. However, the vertical profile of bSOA is also governed by the vapour pressure of the oxygenated intermediates, which decreases with lower temperatures as these intermediates ascend in the troposphere. As they are transported upwards, the reduced volatility at lower temperatures prompts the transition of oxygenated intermediates into the aerosol phase, contributing to the formation of bSOA. Consequently, this interplay between oxidation, transport, and vapour-pressure-driven processes imparts a distinctive D-shaped vertical profile of bSOA concentrations, with elevated concentrations occurring near the surface, followed by an increase in bSOA towards higher altitudes and, ultimately, a decline in the upper atmosphere. At low latitudes (the tropics), this D-shaped bSOA concentration profile is only faint (Fig. 4e). The prevalence of warm air and relatively high boundary layers, as well as the occurrence of deep convection transporting both aerosol particles and precursors into the upper troposphere, leads to maximum gas-to-particle partitioning occurring at higher altitudes, typically around 100–200 hPa, resulting in a secondary local upper-tropospheric enhancement.

Several studies estimate the global mean burden of bSOA to be ∼ 0.5–0.77 Tg . This indicates that, in the present-day land cover scenario, ORACLE provides a lower estimate of the bSOA burden (0.40 Tg) compared to values observed in the literature. evaluated OA values from ORACLE in EMAC and found that surface concentrations are well represented; however, OA is strongly underestimated in the free troposphere. This work by employed the Mainz Organic Mechanism (MOM), which is a more complex chemical mechanism compared to the one used in this study (i.e. the Mainz Isoprene Mechanism – MIM). Nevertheless, this indicates that even with a more complex mechanism, EMAC calculates lower OA values globally. report a 13 % decrease in the global annual mean tropospheric burden of bSOA between 1850 and 2000 due to land use change, while estimate a 91 % decrease in SOA resulting from simulated global deforestation. Our model calculations suggest a reduction of approximately 30 % in bSOA as a result of cropland and grazing-land deforestation compared to natural vegetation.

Regionally, the derived total AOT exhibits opposing effects. In particular, pockets of increased total AOT are evident in confined regions, notably in Southeast (SE) Asia (Fig. 5e). Here, extinction due to H₂O and WASO compounds is found to increase, effectively masking the expected reduction in OA extinction from the lower number of BVOC precursors in the DCGL scenario. This phenomenon is linked to the growth of aerosol particles into what is commonly referred to as the “Greenfield gap” – a range characterised by lower deposition velocities and scavenging values . The extended atmospheric lifetime of particles in these regions adds to the burden of WASO compounds and their associated water uptake, thereby amplifying aerosol extinction. Moreover, the absence of organics exacerbates this effect, leading to disproportionate condensation on accumulation-mode particles at the expense of Aitken-mode particles. Consequently, the increased presence of water-soluble compounds, increased aerosol water content, and shift in size distribution towards larger particles collectively contribute to the observed increase in extinction in these specific regions. While all of these factors contribute, enhanced scattering is the dominant influence on aerosol optical thickness. Nevertheless, plotting the relative difference in total AOT (Fig. S6b) shows that this effect is not very prominent, with an increase in AOT in these regions of less than 4 %. This phenomenon occurs only in the DCGL-PNV scenario. In the DCL-DCGL scenario, the increase in organic aerosols leads to condensation on Aitken-mode particles, limiting growth in the accumulation mode, which aligns with our explanation.

Studies focusing on changes in cloud properties due to perturbed BVOC precursors and SOA loading resulting from land use change are somewhat limited. investigated the impact of bSOA on surface CCN and CDNC at cloud height (approximately 900 hPa) in the present-day atmosphere and found that bSOA increases the annual mean concentration of CCN by 3.6 %–21.1 % and the global annual mean concentration of CDNC by 1.9 %–5.2 %. In this work, we show that present-day deforestation, when compared to the PNV scenario, yields a decrease of 4.8 % in the total column CCN but only a small decrease of 0.2 % in the column CDNC burden, with a slightly more prominent reduction of 0.6 % over tropical regions. The most significant change in cloud droplet count occurs over the Amazon, with an approximate 8 % decrease in this region (Fig. S7d). The changes in CDNC simulated by EMAC, however, seem to be very localised, particularly over the Amazon rainforest, but the global impact on CDNC is small. Deforestation-induced changes in CDNC are less than 10 % over the Amazon and less than 1 % globally (Figs. S6 and S7).

It is worth pointing out that the effect on cloud droplet numbers reported here only stems from aerosol–cloud interactions. In reality, reduced tree cover may lead to less evapotranspiration, which modifies the Bowen ratio (the ratio of sensible heat to latent heat) and potentially influences the cloud effect; however, this feedback is suppressed in our simulations. This is in addition to roughness changes and albedo changes that can also influence cloud cover . Moreover, in this model setup, only aerosol–cloud interactions with large-scale clouds are considered. Therefore, the impact of changes in the aerosol burden on convective clouds and the corresponding potential radiative effects are not captured. The lack of organic NPF descriptions in the model may also result in the underestimation of the real influence of bSOA on CDNC and RE_aci reported in this work.

Biogenic SOA is known to exert net cooling effects on the Earth's climate by scattering a portion of incoming solar radiation back into space . Therefore, changes in aerosol number and composition due to deforestation result in net warming from the lack of aerosol scattering, while reforestation leads to the opposite effect, with a net cooling effect (Figs. 7a and 9e). showed that full deforestation results in a radiative effect of 117 mW m⁻² due to aerosols and a radiative effect of 200 mW m⁻² due to clouds. In contrast, projections for deforestation under the Representative Concentration Pathway (RCP) 8.5 scenario for 2100 indicate that aerosols account for only 6 mW m⁻² of radiative effects, with negligible aerosol–radiation interactions (ARIs) . Furthermore, estimated the total effect of secondary organic aerosol (SOA) aerosol–cloud interactions (ACIs) to be 310 mW m⁻².

The comparatively small ARI effect reported here from the deforestation scenario (60.4 mW m⁻²), relative to the PNV scenario, may stem from differences in methodology compared to other studies, many of which focus on the effects of full deforestation or total SOA (not bSOA). However, our RE_ari estimates might underestimate the actual impact due to comparatively lower bSOA yields in our model runs. The radiative forcing from BVOCs is strongly influenced by SOA yields, which can vary considerably with environmental conditions. SOA yields are sensitive to factors such as temperature and oxidant concentrations, which can alter the overall aerosol burden and the yields' radiative properties . Furthermore, the role of oxidant chemistry is crucial in determining the oxidation pathways and the formation of SOA. Variations in oxidant levels, particularly with respect to ozone and hydroxyl radicals, can substantially affect SOA formation rates and chemical composition . A more comprehensive analysis of the chemical processes influencing SOA yields and their associated radiative forcing should be further explored using the EMAC model.

Regarding the afforestation scenario, we acknowledge that this is an idealistic sensitivity analysis that perturbs only land use cover while maintaining present-day values for greenhouse gas concentrations and anthropogenic emissions. In different future scenarios, these key variables would play a crucial role in determining aerosol radiative effects. This is also highlighted in this study, where we showed that over Southeast Asia, where anthropogenic aerosols are present in relatively high numbers, the reduction in global OA due to deforestation resulted in changes in the aerosol size distribution over this area, leading to an opposing effect in aerosol extinction.

This work includes an analysis of the statistical significance of changes in atmospheric states resulting from perturbed surface BVOC emissions due to land use change. In this analysis, we applied a two-tailed Student's t test with a 90 % confidence level (p<0.1). The t test was performed on annual means to minimise noise resulting from annual internal variability. We show that changes in surface BVOC emissions lead to a statistically significant bSOA column mass burden on a global scale, excluding a portion of the tropical Pacific Ocean. Changes in AOT attributed to OA and total aerosol exhibit statistically significant differences overall, except in Southeast Asia. CCN changes predominantly occur in the tropics (excluding Southeast Asia) and the Southern Ocean. Changes in the CDNC column burden are largely not statistically significant, but CDNC changes at approximately 700 hPa do exhibit statistically significant differences. While RE_ari exhibits a statistically significant signal over the tropical Atlantic Ocean, where most of the changes occur, RE_aci does not demonstrate statistical significance with respect to deforestation.

While some studies, such as those employing the UK Earth System Model (UKESM) and version 2 of the Community Earth System Model (CESM2), indicate statistically significant ARI effects resulting from deforestation at a 95 % confidence level , other studies, such as the one by , employing the Yale-E2 global carbon–chemistry–climate model, suggest that the global-scale impacts of cropland expansion on bSOA are not statistically significant relative to interannual variability in climate. We argue that the land cover changes assessed in EMAC significantly impact global aerosol budgets. However, it appears that the influence of total cloud droplet numbers and RE_aci on a global scale is likely to be minor. Here, we emphasise that the statistical significance of a result is not a property of an effect (whether it is large or small) but rather a measure of whether it can be detected (i.e. how likely it is that the result occurred by chance). Our results suggest that the impact on clouds and the corresponding radiative forcing is not significant; however, this is highly dependent on the model setup. We are limited by a 10-year simulation output as this output may not be sufficient for a signal in clouds to emerge, given the natural noise and internal variability in the climate system.