The Effect of Land Use Classification on the Gas‐Phase and Particle Composition of the Troposphere: Tree Species Versus Forest Type Information

Relationships between vegetation and air quality are intricate and still not fully understood. For regional air quality assessments, a better understanding of the diverse feedback mechanisms is crucial. The present article investigates the impact of land use data set detailedness on air quality predictions. Therefore, two different land use data sets were applied for simulations with COSMO‐MUSCAT for Germany in May 2014. One data set includes detailed information about tree species while the second one obtains generalized widely applied land use classes including mixed and coniferous forests. Moreover, we examined the role of agricultural NO soil emissions, agricultural biomass density enhancements, and model resolution. For a more comprehensive implementation of the secondary organic aerosol (SOA) formation, the SOA module was extended considering additional biogenic volatile organic compound (BVOC) precursor groups from isoprene, α‐pinene, limonene, and sesquiterpene oxidations. The model studies showed substantial differences in BVOC emission patterns between the two land use data sets. The application of detailed tree species information leads to complex BVOC emission patterns with high emission spots. In contrast, coarser forest information lead to standardized comprehensive emissions which result in 50% higher BVOC emissions. These differences affect both the atmospheric oxidizing potential and the production rates of SOA precursors. Land use induced regional differences (tree species minus forest information) in NOx (±2.5%), ozone (−2.5%), OH (±50%), NO3 radical (+70%) concentrations, and SOA (−60%) mass are modeled. Overall, the simulations demonstrate that detailed land use information, extended organic chemistry treatment, and high spatial resolution are mandatory for air quality assessments.

troposphere. The effects of BVOCs on air quality are much more complex and could have both a positive or negative impact (Fitzky et al., 2019;Popkin, 2019). The oxidation of BVOCs regardless of the oxidant can lower their volatility, which enables them to condense onto existing aerosol particles (Odum et al., 1996;Pankow, 1994) or even drive new particle formation (Ehn et al., 2014;Held et al., 2004;Jokinen et al., 2015;Lehtipalo et al., 2018;Riccobono et al., 2014). These two processes result in secondary organic aerosol (SOA) formation contributing to particulate matter (PM), which describes the total aerosol particle mass concentration and is grouped by its aerodynamic diameter (in μm) into PM 1 , PM 2.5 , and PM 10 (von Schneidemesser et al., 2015).
Thus, BVOCs can impair air quality by influencing NO x , ozone, and PM concentrations. So, they can have negative or eventually positive impacts on human health and ecosystems. Unfortunately, BVOC emission estimates bear huge uncertainties (Guenther et al., 2012;von Schneidemesser et al., 2015). BVOC emission models are sensitive to input parameters, such as meteorological conditions (temperature, solar radiation, etc.) and land use cover information (leaf area index LAI, plant functional type, and emission factors). Due to changes in these driving model parameters significant differences in global BVOC emission estimates can be found. Different setups of the widely used global framework MEGAN, the Model of Emissions of Gases and Aerosols from Nature, generate global isoprene emission values ranging from 350 to 800 Tg yr −1 (Guenther et al., 2012). For Europe, a detailed land use data set was published in 2001 containing about 116 tree species (Köble & Seufert, 2001), which is the basis for a comprehensive tree species-specific BVOC emission inventory (Karl, Guenther et al., 2009;Oderbolz et al., 2013;Steinbrecher et al., 2009). The emission algorithms within are based on Guenther et al. (1993) and Guenther (1997) and are therefore based on the same fundamentals as MEGAN. As BVOC emission itself is already highly uncertain, their simplified chemical degradation and approximated SOA forming potential makes climate and air quality assessments challenging. Global SOA budget estimates are still highly uncertain ranging from 12 to 1,820 Tg yr −1 (von Schneidemesser et al., 2015).
BVOC emission strengths and compositions are quite distinct between different tree species (Karl, Guenther et al., 2009;Steinbrecher et al., 2009) leading to varying air quality impacts. Additional land use induced effects on air quality are caused by changes in the meteorological fields primarily related to LU changes (urbanization, renaturation, deforestation, management, etc.) due to climate change or human activities (De Meij et al., 2015;De Meij & Vinuesa, 2014;Findell et al., 2017;García-García et al., 2020;Miralles et al., 2019;Seneviratne et al., 2010). Prescribed LU characteristics, such as the LAI, plant cover, root depth, and LU specific roughness length affect all surface fluxes (emissions, radiation, momentum, sensible, and latent heat) leading to changes in wind speed, temperature, relative humidity, and the boundary layer height pending on the land surface model used (for COSMO the land surface model is TERRA, e.g., Doms et al., 2013Doms et al., , 2018. Furthermore, the handling of the flux calculation affects the model performance. Two approaches are possible: (a) the dominant/bulk and (b) the mosaic approach. While the dominant/bulk approach only considers the dominant LU or bulk (weighted average) LU quantities, the mosaic approach considers individual contributions of all LU categories within the grid cell, which makes this approach less sensitive to spatial resolution (Li et al., 2013). Additionally, the deposition is linked to the LU classification (Schlünzen & Pahl, 1992;Simpson et al., 2012;Zhang et al., 2003).
In order to address climate mitigation and adaptation strategies as well as air quality assessments, more detailed knowledge of tree species-specific quantities are needed as currently most of the vegetation impacts are based on plant functional types. In the present article, we investigate the impact of LU classification detailedness on BVOC emissions by using two different LU data sets. While one data set contains detailed information about 138 LU classes (138_LU) including 116 tree species, the other one consists of 10 LU classes including only 2 simplified forest categories: mixed and coniferous forest (10_LU). Here, the effect of a reduction toward the forest level not only on BVOC emissions but also on the atmospheric oxidizing potential as well as air quality in terms of OH, NO 3 , ozone, NO x , and SOA concentrations is investigated with the model system COSMO-MUSCAT (see Section 2.1) for Germany in May 2014.
To further extent the representation of BVOCs in the chemical mechanism, reactions for sesquiterpene are implemented into the applied chemistry mechanism RACM (Stockwell et al., 1997). Furthermore, SOA products from monoterpenes + NO 3 reactions as well as isoprene and sesquiterpenes are added to the SOA module SORGAM (Schell et al., 2001). Several simulations were carried out for different scenarios and were compared with measurements from the field site Melpitz.

The Model System COSMO-MUSCAT
The MUltiScale Chemistry Aerosol Transport model MUSCAT is a state-of-the-art chemical transport model developed at the Leibniz Institute for Tropospheric Research (TROPOS). MUSCAT allows the use of several modules for emissions, chemistry, aerosol formation, and dynamics (Wolke et al., 2004(Wolke et al., , 2012; https://www.tropos. de/en/research/projects-infrastructures-technology/technology-at-tropos/numerical-modeling/cosmo-muscat). COSMO (version 5.05) is the numerical weather prediction model of the German weather service DWD (Schättler et al., 2018) and is online coupled to MUSCAT and works as the meteorological driver. The good performance of COSMO-MUSCAT regarding air quality simulations has been validated and confirmed by model intercomparison studies, for example, the Air Quality evaluation International Initiative AQMEII (Im et al., 2014a(Im et al., , 2014bSolazzo, Bianconi, Pirovano et al., 2012;. The studied model area is Germany (N2; orange area in Figure S1 in Supporting Information S1). To generate the initial boundary conditions for the model domain N2 a European model run (N1; brown area in Figure S1 in Supporting Information S1) has been performed first. The European run itself uses 3 hr ECMWF-IFS CAMS boundary and initial data for aerosols (sea salt, dust, organic matter, black carbon, and sulfate), gas-phase concentrations (NO 2 , CO, SO 2 , HCHO, H 2 O 2 , O 3 , NO, HNO 3 , OH radicals, PAN, isoprene, methane, methanol, ethene, ethane, ethanol, and propane), and temperature data. The resolution of the model domain N1 is 28 km × 28 km and for N2 7 km × 7 km and roughly 22 km height for COSMO for both domains while N1 contains 40 and N2 50 levels; 10 and 12 levels being within the first 1,000 m, respectively, with the first level at 20 m. MUSCAT is restricted to 8 km height. MUSCAT uses a multi-block approach which allows different horizontal resolutions by defining multiple blocks within the grid (Knoth & Wolke, 1998a, 1998bWolke & Knoth, 2000). An implicit-explicit (IMEX) time integration scheme is used in MUSCAT (Knoth & Wolke, 1998a, 1998bWolke & Knoth, 2000). The explicit time step is based on the slow process of horizontal advection and the corresponding CFL-criterion, while the implicit time steps take into account the faster processes of vertical exchange and chemical transformation splitting one explicit time step into several implicit ones separately for every block.

Land Use Data Sets
Two simulations with different LU data sets, the 138_LU (Köble & Seufert, 2001) and 10_LU are carried out to test the sensitivity of the model toward tree species versus forest level information. The 138_LU contains 116 tree classes, 11 agricultural land use types, and additional vegetated (wetlands, grassland, green urban area, etc.) and non-vegetated land use categories (water bodies, artificial surfaces, etc.). To study the effect of more generalized land use types, commonly used for global model estimates, the 138 land use classes have been transformed into the following 10 classes: water, mudflats, sand, mixed land use (mainly agriculture), meadows, heath, bushes, mixed forest (broadleaved trees), conifer forest (coniferous trees), and urban area. In this way, the contribution of trees and forests in a grid cell are kept equal. These 10 classes have been used in COSMO-MUSCAT simulations for this model domain so far and are linked to the implemented deposition scheme (Schlünzen & Katzfey, 2003;Schlünzen & Pahl, 1992). The simulations using the generalized land use will have the prefix 10_LU. The data set from Köble and Seufert (2001) has a resolution of 1 km × 1 km. For the simulations, the data set has been resampled onto the model grid giving the percentage distribution of all land use classes for each grid cell (Figure 1).
The distribution of the different LU categories is crucial for the BVOC emission patterns which subsequently affect the spatial and temporal distribution of oxidizing agents and reaction products. In preparation for the model results the most common vegetation types of both LU data sets are addressed first. The Figures 1a, 1e,and 1i belong to the 10_LU data set, Figure 1d is part of both data sets and the remaining panels show the distribution for the 138_LU. If more than one tree species is present in the model domain for the 138_LU they are grouped together and presented on a genus level. The first row depicts agricultural LU categories, the second row deciduous and the third-row coniferous tree classes. Comparing the general LU class mixed land use (Figure 1a) with Figure 1b reveals non-irrigated arable land as the dominating agricultural LU category and is present throughout the whole domain with a large proportion from the Netherlands through Germany with a maximum in the center till the Czech Republic which is here denoted as the NW-SE agricultural band (see Figure S1 in Supporting Information S1). The LU class representing agricultural and natural vegetation (Figure 1c) is more pronounced in the Czech Republic, in Poland (line parallel to the Baltic Sea) and a spot in the Netherlands close to the German border. Meadows (Figure 1d) are dominating along the coast of the North Sea. The agricultural LU categories are of importance for the sensitivity studies presented in Section 3.4 and 3.5.
All deciduous trees are grouped into the forest category mixed forest (Figure 1e) for the 10_LU and occur more frequent in the middle and south west of the domain. For 138_LU oak ( Figure 1f: Quercus robur, Quercus petraea, and Quercus rubra) and birch (Figure 1g: Betula pubescens and Betula pendula) species as well as common beech (Figure 1h) are the most important broad-leafed tree species for Germany. If more than one tree species is present the predominant one is underlined. Common beech is the most frequent occurring broad-leafed tree species followed by oaks. Beech show higher ratios in the middle and south west of the domain, additional spots appear in northern Poland and Germany. Oaks are present in small rations throughout the whole domain but larger fractions occur around Berlin, a few spots in Poland and more frequent in the west of the domain. Smaller spots can be found for birch species in the whole domain with largest proportion in Poland.
The generalized LU category conifer forest (Figure 1i) sums up all the coniferous tree species and shows a higher percentage distribution in the east and south of the domain. When considering tree species, a clear separation between pine (Figure 1j: Pinus sylvestris, Pinus nigra, and Pinus strobus) and spruce (Figure 1k: Picea abies and Picea sitchensis) species is visible. Pine species dominate the north east (Brandenburg and Poland). They also reach around from Brandenburg to the Netherlands and along the German border until the Rhine and are frequently present in the south. Spruce species predominate the lower mountain range which is in the southern model domain and a few areas in the north for which the highest ratio is reached in north Poland. Douglas fir ( Figure 1l) show only minor overall contribution but its huge BD (of 1,000 g m −2 ) makes it an intriguing tree species when looking at the effects of BVOC emissions on air quality. Only a few spots are present in the north, west and one in the Czech Republic with the highest contribution in the Netherlands.

Emissions
For anthropogenic emissions, the MACC 2010 (Kuenen et al., 2014;Pouliot et al., 2015) inventory is applied for Europe. For Germany, data from the federal environmental agency (UBA) was made available using GRETA, the Gridding Emission Tool for ArcGIS (Schneider et al., 2016). BVOC emissions E sl are parametrized according to Steinbrecher et al. (2009) And uses information about the area A covered by a certain LU type (indicated by lower index l), biomass densities (BDs in g m −2 ), plant species-specific standard emission potentials SEPs (μg g −1 h −1 at 30°C and photosynthesis-active radiation of 1,000 μmol m −2 s −1 ) for varies BVOCs (indicated by lower index s) and a correction factor γ for seasonality, temperature, and light. A more detailed discussion of BVOC emissions parametrization and its constraints can be found in the supplement (Text S1 in Supporting Information S8). The product of BD with the SEP is called the standard emission factor. The following BVOC emissions are included: isoprene, 17 monoterpenes (α-and β-pinene, camphene, Δ 3 -carene, sabinene, α-thujene, limonene, α-and γ-terpinene, α-and β-phellandrene, myrcene, trans-and cis-ocimene, linalool, p-cymene, and 1,8-cineol), sesquiterpene (β-caryophyllene), and oxygen containing BVOCs (OVOCs: methanol, acetone, ethanol, acetaldehyde, formaldehyde, formic acid, and acetic acid). Trees are supposed to emit sesquiterpenes while for other land use categories, such as agriculture and grass land it remains uncertain. Karl, Guenther et al. (2009) suggested sesquiterpene emissions for all LU categories including agriculture while Steinbrecher et al. (2009) only present sesquiterpene SEPs for trees but not for agricultural LU categories. Oderbolz et al. (2013) list 0 as SEP for agricultural sesquiterpene emission. Therefore, here crops are treated as non-sesquiterpene emitting and other land use categories (natural and urban vegetation, shrubs, and trees) are assigned 0.1 μg g −1 h −1 . Except for agriculture and a few tree species, all LU classes are assigned a sesquiterpene SEP of 0.1 μg g −1 h -1 ( Karl, Guenther et al., 2009;Oderbolz et al., 2013;Steinbrecher et al., 2009).
BVOCs impact both the production and degradation of radicals and photo oxidants. Therefore, a simplified description of the implemented VOC chemistry is given in the following, only showing reactions used in the chemistry mechanism RACM-MIM2-ext. A more complex VOC chemistry description can be found in Atkinson (2000) and Atkinson and Arey (2003 In the presence of NO x ozone is quickly oxidized by NO (R5) producing NO 2 which gets photolyzed (R6) to NO and O( 3 P). Between these reaction sequence, a photostationary state is established (Leighton relationship).
This equilibrium is disturbed by BVOC degradation processes through the production of intermediate BVOC products (R7), the organic peroxy radicals (RO 2 ). Instead of ozone, NO can also react with RO 2 to form NO 2 (R8). In addition, HO 2 is produced which can also react with NO to form NO 2 and OH (R9). Besides the reaction with NO, RO 2 radicals can react with other radicals: HO 2 , methyl peroxy (MO 2 ), acetyl peroxy (ACO 3 ), NO 3 , and self-reactions (not shown). Furthermore, oxidized BVOCs are produced from R7 and R8 and subsequent RO 2 oxidation processes so that R7 initializes SOA formation. Additional SOA initialization reactions of BVOC equivalent to R7 are the reaction with ozone or the nitrate radical (NO 3 ; both not shown).
Nitrate radicals are produced by reactions of NO 2 with ozone (R10). During the day, NO 3 radicals are rapidly photolyzed (R11a and R11b), or quickly react with NO (R12). The reaction of OH radicals with NO 2 leads to the formation of nitric acid (HNO 3 ) and is a major daytime NO x and HO x (sum of OH and HO 2 ) loss processes (R13). During night, the formation of dinitrogen pentoxide (N 2 O 5 ) from the reaction of NO 3 radicals with NO 2 , which is a reversible process (R14), and its wet and dry deposition as well as aerosol uptake are important NO x loss processes due to the hydrolysis into HNO 3 . The production of nitrous acid (HONO) from OH radicals and NO (R15) is in photo equilibrium with HONO photolysis (R16).
BVOCs can be quickly oxidized by ozone, OH, or NO 3 radicals (see lifetime Figures S42-S44 in Supporting Information S7; Atkinson, 2000;Atkinson & Arey, 2003;Oderbolz et al., 2013). OVOCs, p-cymene, and 1,8-cineol react preferably with OH radicals (Atkinson & Arey, 2003;Corchnoy & Atkinson, 1990;Oderbolz et al., 2013). Isoprene can react fast with OH radicals and somewhat slower with NO 3 radicals and O 3 (Atkinson & Arey, 2003;Oderbolz et al., 2013). Monoterpene groups API and LIM are oxidized by OH radicals during the day and rapidly by NO 3 radicals during night, reactions with O 3 can also occur throughout the day (Atkinson & Arey, 2003;Oderbolz et al., 2013). Degradation of sesquiterpenes is dominated by O 3 followed by NO 3 and OH radicals (Atkinson & Arey, 2003;Oderbolz et al., 2013). The volatility of anthropogenic (AVOC) and BVOCs drops with every chemical functionalization process enabling the oxidation products formed to condense onto particles (Odum et al., 1996;Pankow, 1994) or initiate new particle formation (Ehn et al., 2014;Held et al., 2004;Jokinen et al., 2015;Lehtipalo et al., 2018;Riccobono et al., 2014) and thereby producing SOA. The SOA yield itself depends on the specific VOC oxidant reaction and the total available absorbing organic matter OM (Pankow, 1994). The initial reaction of sesquiterpenes with ozone followed by a second oxidation leads to a SOA yield of more than 100% (Hoffmann et al., 1997) making it the most efficient one followed by the reaction of monoterpenes and isoprene with NO 3 , O 3 , and OH, respectively.

Updated SOA Module
The formation of SOA is described by the 2-product approach SORGAM (Schell et al., 2001) which was implemented into RACM-MIM2-ext without changing the gas-phase chemistry mechanism. The original SORGAM module only includes SOA products for API and LIM from OH and ozone oxidation, as well as anthropogenic VOCs: aromatic precursors (toluene, xylene, and cresol), higher alkane (HC8), and higher alkenes (OLT and OLI). During the plant growing season and BVOC emission favoring meteorological conditions (warm temperatures and high solar radiation) SOA production is substantially underestimated mainly due to lacking BVOC SOA treatment (see Section 3.7). To improve the SOA formation model performance, SORGAM was updated by reactions for monoterpenes with NO 3 (Griffin et al., 1999), isoprene (Kroll et al., 2006;Ng et al., 2008), sesquiterpenes (Hoffmann et al., 1997;Karl, Tsigaridis, Vignati, & Dentener, 2009), and reaction pathways for highly oxidized multifunctional organic molecules HOMs (monoterpene: Berndt, Richters et al., 2016;Jokinen et al., 2015;sesquiterpenes: Richters et al., 2016;isoprene: Berndt, Herrmann, Sipilä, & Kulmala, 2016). The BVOC SOA products and corresponding reactions are summarized in Tables S1 and S2 in Supporting Information S1. For monoterpenes, the saturation vapor pressure (P 0 ) values from Schell et al. (2001) were used while the stoichiometric coefficients α i were calculated with python3 from SOA yield measurements for β-pinene NO 3 radical reaction from Griffin et al. (1999) showing a correlation of 0.9963. Differences in α i for API and LIM are due to different molar masses of the SOA products. The values for P 0 of isoprene are chosen in a way that the fit for OH as well as NO 3 radical reactions in combination with the corresponding α i show good correlations with SOA experiments (see Figure S55 in Supporting Information S8; OH: Kroll et al., 2006;NO 3 : Ng et al., 2008) while prioritizing the NO 3 radical reaction due to its higher SOA forming potential (0.906 and 0.998, respectively). For clear separation of biogenic and anthropogenic SOA sources as well as their oxidizing agents, all SOA products are addressed separately resulting in 38 different SOA products (see Table S1 in Supporting Information S1).

Melpitz
To test the SOA module update and compare the simulations with measurements, the time period May 2014 was chosen as during this month an intensive measurement campaign took place at the TROPOS field site Melpitz (see Section 3.7). The TROPOS research site Melpitz (51.54°N, 12.93°E, 86 m a.s.l) is a regional background station situated about 50 km to the Northeast of Leipzig in a glacial valley of the river Elbe. The research site Melpitz itself is on a flat meadow surrounded by agriculture (Spindler et al., 2013). The grid cell within COSMO-MUSCAT that includes Melpitz consists of 87.7% agriculture, 7.9% artificial surfaces, 2.6% pine forests, and 1.8% water bodies. Melpitz is integrated in EMEP (Co-operative program for monitoring and evaluation of the long-range transmission of air pollution in Europe), as well as the European scientific infrastructure ACTRIS (Aerosol, Clouds and Trace gases Research Infrastructure) and the German Ultrafine Aerosol Network

Model Results
The influence of LU data set detailedness on air quality is addressed by difference plots of the average map plots during May 2014 for various scenarios between the detailed and the generalized LU data set focusing on BVOC emissions induced responses to tree species versus forest classifications. The differences between the two LU data sets are key to model simulation analysis and are therefore continuously shown, always comparing 138_LU with 10_LU (138_LU-10_LU), if not stated otherwise. The diversity of LU classes and their individual SEP and BD for the 138_LU strongly influences BVOC emission strengths, composition, and individual patterns. Note that the tree species composition is crucial for the BVOC composition and therefore, the results of this regional analysis cannot directly be transferred to other regions with different tree populations. An overview of minimum, average, and maximum values of the different scenarios can be found in Table S3 in Supporting Information S1.
While analyzing the influence of detailed LU versus simplified forest information on BVOC emissions the link between certain tree species and corresponding low and/or high BVOC emissions as well as oxidants, BVOCs, and SOA concentration levels are investigated. Additionally, the impact of NO on these interdependencies are addressed by secondary agricultural NO soil emissions. For the same reason, agricultural BD is increased which affects the BVOC emission strengths. Next to these model sensitivity studies the model spatial and temporal resolution was refined as preliminary step for comparisons with measurements from Melpitz.

BVOC Emission Differences
The two LU data sets have a major impact on BVOC emissions and can cause drastic regional changes. For the 138_LU isoprene emission ( Figure S2 in . Except for the tree species spots the difference between the LU data sets is always negative (Figure 2 and Figure S2 in Supporting Information S2).
Monoterpene emission is governed by coniferous species of pine and spruce trees (Figures 1j and 1k) and to some extend also by beech (Fagus sylvatica; Figure 1h). Table S4 in Supporting Information S1 includes values for BD, SEP, and SEF for the four monoterpene groups α-pinene, limonene, p-cymene, and 1,8-cineol for the LU classes Picea abies, Pinus sylvestris, Fagus sylvatica, coniferous, and mixed forest. The two forest categories include the standard monoterpene emission split. According to Steinbrecher et al. (2009), 32 tree species show an individual monoterpene emission split based on measurements including the two dominating conifers Picea abies and Pinus sylvestris as well as beech. The remaining tree species were assigned a default monoterpene split (API 71%, LIM 23%, BARO 4%, and BHC8 2%).
Picea abies predominates the lower mountain range of Germany while Pinus sylvestris is more frequent in the north east. The combination of SEF with the emission split results in similar but lower α-pinene emission for pine, spruce, and most deciduous trees. Beech has a significant higher SEF. These tree species induced differences in SEF compared to the standardized values cause alternating emission values in the middle and south west of the domain as well as northern Poland while the rest is negative (Figure 2 and Figure S3 in Supporting Information S2). Limonene emissions are lower throughout the whole model domain (Figure 2 and Figure S3 in Supporting Information S2) especially for pine forests. The main contributor to 1,8-cineol emission ( Figure  S4 in Supporting Information S2) is Picea abies, followed by birch trees and again a decrease for pine forests ( Figure 2). p-Cymene are emitted by Pinus nigra and Douglas fir, which not frequently occur in the model domain. Therefore, p-cymene emission is highly overestimated (six times higher mean value; see Table S3 in Supporting Information S1) when using 10_LU ( Figure S4 in Supporting Information S2). Note that the default monoterpene emission split is used for Pinus nigra and Douglas fir. Pending on the literature used, both could be treated as non p-cymene emitters or with low contribution to the monoterpene emission. Steinbrecher et al. (2009) reports a contribution of 4% for p-cymene while Geron et al. (2000) and Faiola et al. (2015) state a contribution of 0% and 1.7% for Douglas fir. Pinus nigra is identified as non p-cymene emitter (Geron et al., 2000;Harley et al., 2014) but small concentrations were detected in essential oils from needles, branches, and cones (Macchioni et al., 2003;Supuka et al., 1997).
Sesquiterpene ( Figure S4 in Supporting Information S2) and OVOC ( Figure S2 in Supporting Information S2) emissions are almost exclusively controlled by BD as most tree species have the same SEP (sesquiterpenes: 0.1 μg g −1 h −1 ; OVOC: 2 μg g −1 h −1 ). Exceptions are birch trees (SEP: 2 μg g −1 h −1 ) for sesquiterpenes, and spruces (Picea abies 2.3 μg g −1 h −1 ; P. sitchensis: 2.89 μg g −1 h −1 ) as well as beeches (10 μg g −1 h −1 ) for OVOCs. The smaller BD of pine trees generates lower and the higher BD of spruces higher OVOC and sesquiterpene emissions. Higher sesquiterpene emissions are also located in the southern model domain and Poland's birch and spruce forests ( Figure S4 in Supporting Information S2). For OVOCs, higher emissions occur in the German lower mountain ranges and Poland's spruce and beech forests (Figure 2 and Figure S2 in Supporting Information S2). Higher OVOC emissions over the Netherlands are caused by a larger BD of the agricultural LU class complex cultivated patterns (distribution not shown) compared to the mixed LU class of 10_LU. As agricultural LU categories are treated as non-sesquiterpene emitting this is not evident in the sesquiterpene emission plot.
These results are in good agreement with Steinbrecher et al. (2009) and Oderbolz et al. (2013) which are both based on the LU data set from Köble and Seufert (2001). They present similar emission patterns but for time periods July 2000 and June 2006, respectively. The emissions are in the same range (0-350 kg km −1 month −1 ), however, differences are related to different model systems, meteorological conditions, and time of the year. Unfortunately, these articles only focused on BVOC emissions.
The reduction to forest level information leads to 50% overestimated BVOC emissions (excluding OVOCs) compared to detailed trees species information influencing not only BVOC emission strengths but also the emission patterns and composition. Certain tree species emit higher amounts than their corresponding forest category resulting in BVOC emission spots: as every single BVOC has its preferred oxidizing partner, their degradation strongly effects the distribution and concentration of their oxidants: ozone, OH, and NO 3 radicals.

Modeled BVOC Concentration Patterns
The mean modeled BVOC concentration patterns are similar to the emissions mainly just with changed units and scales. Therefore, they are not described in detail here but are given for the sake of completeness in the supplement (see Figures S5-S11 in Supporting Information S2).

Link Between BVOC and Oxidant Concentrations
LU induced BVOC emission changes affect the tropospheric oxidants budget by consumption due to degradation as well as production processes. For OH and NO 3 radicals, the consumption dominates, always linking higher BVOC emissions with lower oxidant concentrations in the grid cell. For ozone, it is more complex due the overlaying consumption and production processes which are additionally linked to RO 2 and NO x concentrations (mainly R1-R10). In order to examine certain pattern or possible interplays between BVOC concentrations and oxidants, not only difference plots but also temporal correlation analyses for the entire grid were performed to gain additional information. Temporal correlations for every single grid cell are calculated using the Pearson sample linear cross-correlation function for ground-level model concentration values. Instead of showing several time series for numerous spots of the model domain and calculating correlations individually, here the correlation between two time series of different concentrations in every single grid cell of the domain are summarized in correlation map plots (Figures S13, S14, S16, S17, S19, S20, S24, and S25 in Supporting Information S3 and S4). A few sets of time series with additional information for selected spots are given in the supplement ( Figure  S45-S52 in Supporting Information S7). However, the interpretation of correlations between BVOCs and other concentrations are intricate as the emission of BVOCs, their chemical degradation as well as oxidant production show either similar or opposing diurnal cycles. Furthermore, their distribution is additionally influenced by the meteorology. BVOC emissions are stimulated be temperature and solar radiation. Synthesis emission onset comes with radiation while pool emission might rise earlier if temperature already increases before sunrise. These enhanced morning emissions are followed by concentration increases. Atmospheric photolysis processes proceed at the same time producing ozone and thus initiate OH radical generation. If oxidant production becomes efficient, the BVOC depletion by chemical reaction with the oxidant compensates the BVOC emission and the BVOC concentration declines until photo-chemical oxidant production becomes inefficient. If BVOC emission progresses, they can accumulate again until in the case of synthesis emission their emission ceases. In this case, if chemical degradation progresses, the BVOC might become totally depleted. For pool emission, lower emissions will continuously provide BVOCs which leads to an accumulation during night if their oxidation is governed by OH radicals. In the case of NO 3 degradation pathways, the same as for the photolysis oxidants repeats. Until NO 3 production via R10 becomes sufficient, BVOCs accumulate. Afterward, BVOC depletion starts until the NO 3 becomes insufficient and an accumulation appears again until photolysis sets in.

OH
The OH mean concentration ( Figure S12 in Supporting Information S2) is primarily linked to isoprene and second to OVOCs reducing OH radical concentration in high emission regions (Figure 2). The difference plots ( Figure 2 and Figures S2, S5, and S12 in Supporting Information S2) are therefore inversely colored showing maximum values for isoprene and at the same location minimum values for OH and vice versa. Except for isoprene emission spots of oaks, isoprene concentration for 138_LU are low reaching nearly 100% reduction in forests predominantly consisting of pines and second spruce. Highest OH concentration rise is reached in non-isoprene emitting pine forests. The higher OVOC and low isoprene emissions (Figure 2) of spruce and beech trees diminish the OH concentration rise and reduce the contrast to the generalized forest categories in the southern model domain.
Main OH production occurs during the day via photochemistry therefore, only day-time values are considered (concentrations between 4 UTC and 18 UTC + 0:00, with the maximum at 12 UTC) for temporal correlations. Lifetime of BVOCs due to OH is in the range of 0.5 till 2.5 hr for API, LIM, ISO, and SQT ( Figure S42 in Supporting Information S7). For isoprene, the 10_LU shows positive correlations ( Figure S13 in Supporting Information S2) as OH production and isoprene emission have similar diurnal cycles and the moderate meteorological conditions till mid of May inhibits an effective early morning rise for synthesis emissions like isoprene and hinder photosynthesis OH production. In the 138_LU correlation plot also negative values are present ( Figure S14 in Supporting Information S2) at high isoprene emission spots of oaks. Here, the fast intensification of isoprene emission with the onset of solar radiation quickly increase isoprene concentration which is then compensated by OH production until it becomes insufficient and isoprene accumulates again ( Figures S51 and S52 in in Supporting Information S7). Negative BVOC correlations are present where the early morning peak is followed by a decrease, which except for the isoprene emission spots is more likely for pool emissions ( Figure S49 in in Supporting Information S7) in BVOC rich environments, as their emission already sets in before sun rise and before sufficient amounts of OH can form. Beech and spruce are mainly defined as synthesis emitters and, therefore, have more similar diurnal cycle with OH ( Figures S46 and S50 in in Supporting Information S7) and show positive correlations with α-pinene and limonene. The same applies for 1,8-cineol which is nearly solely emitted by spruce ( Figure S4 in Supporting Information S2) and also shows a positive correlation. The diurnal cycle for 1,8-cineol concentration of spruce is similar to the emission ( Figure S50 in in Supporting Information S7). Here, the higher emissions exceed the oxidation by OH and the increase during the day is more significant than the accumulation overnight. For 10_LU limonene and α-pinene show almost the same correlation plots ( Figure S13 in Supporting Information S3) as in the standard set up the difference between the two is invoked by the default monoterpene emission split and thus have similar diurnal cycles (see the 10_LU plots of Figures  Sesquiterpenes have a positive impact on OH as their oxidation by ozone leads to OH-radical production (Karl, Tsigaridis, Vignati, & Dentener, 2009). In high NO x environments such as the Ruhr area, this effect is limited by the fast cycling between NO x and ozone (R1 till R6) which creates a strong positive correlation between OH and ozone. OVOCs show slightly positive correlations for coniferous forest/pine and spruce trees and negative values for mixed forest/oaks and beech. The night-time accumulation of conifers pool emissions leads to an intensified consumption of OH in the morning reducing the amount of OH for the slower reaction of OVOC which delays and impairs the onset of depletion (compare Figure  S46 and Figure S47 in Supporting Information S7).

NO 3
The reaction of α-pinene with OH is rather slow and the consumption of OH radicals by isoprene and OVOCs makes the α-pinene NO 3 degradation path even more important (Ng et al., 2017). There is a direct link between α-pinene ( Figures S3 and S6 in Supporting Information S2) and NO 3 concentrations ( Figure S15 in Supporting Information S3). Again, showing invers colors in the difference plots. This is more pronounced for the reduced α-pinene emission of spruce trees, here the α-pinene concentration is roughly halved while at the same time NO 3 concentration doubles.
The rapid photolysis of NO 3 during the day by R11a and R11b keeps daytime NO 3 concentration low. But during night, the concentration builds up and NO 3 becomes an important BVOC oxidant. Therefore, only night-time values are considered (between 19 UTC and 3 UTC) in the correlation calculation (Figures S16 and S17 in Supporting Information S3). Night-time BVOC emissions are rather low or even not existing for synthesis emissions (isoprene, synthesis monoterpene). Additionally, the progressing BVOC degradation further reduces their concentrations during night and thus opposing trends compared to NO 3 can be observed independent of the LU for isoprene, sesquiterpenes, limonene, and α-pinene and thus are negatively correlated. Largest negative values are reached for limonene and α-pinene due to their fast NO 3 oxidation (lifetime of minutes Figure S43 in Supporting Information S7; Oderbolz et al., 2013). 1,8-Cineol, p-cymene and OVOCs are emitted in low quantities during night but as they nearly solely react with OH-radicals ( Figures S42-S44 in Supporting Information S7; Corchnoy & Atkinson, 1990;Oderbolz et al., 2013), their concentration builds up due to lacking oxidants resulting in a positive NO 3 correlation.
However, in areas with high BVOC concentrations a negative NO 3 correlation is calculated, because of significant OH and HO 2 generation by ozonolysis processes. This also results in negative correlation between NO 3 and ozone (Stockwell et al., 1997). Furthermore, the reaction of RO 2 with NO leads to NO 2 and HO 2 production (R8). Subsequent reactions of HO 2 with NO and NO 3 are additional OH sources (Stockwell et al., 1997). Due to the enhanced BVOC emissions of the 10_LU run, a higher OH radical production occurs for coniferous forests compared to the 138_LU run increasing the area of negative correlations of NO 3 with ozone, OVOCs, 1.8-cineol, and p-cymene. Night-time BVOC emissions are considerably reduced for the 138_LU as the most common tree species show higher ratios of synthesis emissions with pine being the main contributor to night-time BVOC emissions. Therefore, negative correlations are only present in pine dominated surroundings. The night-time production of OH is visible in the time series, mainly for spruce and pine, especially at the second half of May where higher temperatures and radiation amplify BVOC emissions ( Figures S45-S52 in Supporting Information S7).
Here, OH radicals are not completely degraded during night. The synthesis emissions of beech inhibit night-time ozonolysis thus no significant OH production occurs.

Ozone
The ozone concentration budget ( Figure S18 in Supporting Information S3) strongly depends on NO x (R4, R5, and R6) and VOC levels as well as on VOC composition (R7, R8, and R9). Differences in ozone are mainly linked to isoprene and limonene emission modification as they changed the most between the two LU data sets ( Figures  S2, S3 in Supporting Information S2 and Figure S18 in Supporting Information S3). The most considerable ozone reduction occurs in pine forests in east Germany (up to 2.5%). In the western domain, similar ozone values are modeled due to intensified isoprene emissions of oaks.
The 10_LU shows higher positive correlation values ( Figure S19 in Supporting Information S3) than the 138_LU ( Figure S20 in Supporting Information S3) with a north-east south-west gradient. The lifetime of BVOCs due to ozone is in the range of a few minutes for β-caryophyllene, hours for API and LIM, 1 day for isoprene and more than 100 days till years for 1,8-cineol, p-cymene, and OVOCs ( Figure S44 in Supporting Information S7; Atkinson & Arey, 2003; Oderbolz et al., 2013). Therefore, a direct link to tree species is difficult as here not just the ozone consumption but also the production through other reaction pathways is tedious. Both processes overlap each other and are additionally impacted by transport processes. This makes the correlation even more sensitive to the diurnal cycle. Oaks and spruces show higher positive isoprene ozone correlation values whereas areas with lower correlation values show elevated fractions of birch, beech, and pine trees, which are all classified as non-isoprene emitters (Figure 1 and Figure S20 in Supporting Information S3). Monoterpenes, sesquiterpenes and OVOCs show negative correlations for high NO x environments (Ruhr area, Berlin). Here, the reduction of ozone, OH-(R1, R2, and R3), and NO 3 -(R10) radical concentration lowers the oxidizing ability which increases reactant concentrations. Sesquiterpenes are most sensitive to the NO x -induced ozone depletion as they are preferably oxidized by ozone (Oderbolz et al., 2013). The plots for the 138_LU ( Figure S20 in Supporting Information S3) depict more substantial regional differences in the correlations for 1,8-cineol and p-cymene. Both are not emitted by beech and pine, while spruce emit higher quantities generating higher positive correlation values. p-Cymene is also not emitted by spruces. Just a few tree species such as Douglas fir and Pinus nigra emit sufficient amounts of p-cymene mainly resulting in positive correlations. Oaks emit minute quantities of p-cymene which makes it an unlikely reaction partner and accumulation occurs more frequently resulting in opposing diurnal concentration cycle ( Figure S52 in Supporting Information S7) and cause negative correlations in the southwest model domain. These effects associated with 1,8-cineol and p-cymene might change when exchanging the default emission split to measurement-based values.
Anthropogenic emissions are the same for both simulations but LU induced BVOC emission deviations affect NO x concentrations. Lower BVOC emissions of the 138_LU especially of pine and spruce forests reduce the amount of organic peroxy radicals RO 2 (R7) and limits their NO degradation channel (R8) lowering HO 2 and NO 2 production and increase NO concentration. However, the RO 2 reduction promotes the oxidation of O 3 by NO (R5) which reduces O 3 while increasing NO 2 concentration. Therefore, HO 2 and ozone concentration is reduced whereas NO and NO 2 concentration is slightly increased for the 138_LU (Figure 3). The increase in NO 2 in combination with increased OH concentration of pine forests facilitates HNO 3 production (R13) with a maximum increase in Brandenburg (Figure 3). In the western model domain, similar NO and ozone values are observed. Here, the frequent occurrence of larger cities generates higher NO x ( Figures S21 and S22 in Supporting Information S3) but lower ozone ( Figure S18 in Supporting Information S3) concentration as well as lower OH concentrations ( Figure S12 in Supporting Information S3) due to oaks/mixed forests. The increase in NO 3 -radical concentration for spruce trees due to their lower α-pinene emissions enables the formation of N 2 O 5 (R14, Figure 3) which leads to NO 2 removal (blue colored bow around Luxembourg Figure 3). The highest NO increase (Figure 3 and Figure S21 in Supporting Information S3) can be observed for larger cities (e.g., Ruhr area and Frankfurt am Main) and landscapes shaped by surface mining (East Germany). In high NO x environments, the increased BVOC emissions of the 10_LU amplifies NO 2 production (R8) creating the negative NO 2 differences which are more pronounced in BVOC rich environments in East Germany (Figure 3 and Figure S22 in Supporting Information S3).
NO and NO 2 show negatively ozone correlations (Figures S19 and S20 in Supporting Information S3). The early morning NO emission peak quickly reduces ozone and produces NO 2 (R5). If ozone is completely depleted, NO concentration rises again but decreases when photolytic ozone production becomes efficient (R6 and R4). Due to the high day-time ozone values, NO afternoon emissions do not manifest in NO but in a NO 2 concentration rise while at the same time ozone decreases. Therefore, negative correlations are more pronounced for NO 2 . Largest negative correlation values for both NO and NO 2 occur in high NO x environments. Additional ozone production results from the reaction of RO 2 by NO molecules (R8 followed by R9, R6, and R4). Hence, lower ozone concentration combined with lower BVOC concentration indicate a production dependency rather than a consumption as for the other oxidants.

Impacts on SOA Production
Tropospheric oxidation of BVOCs leads to the formation of low volatile reaction products enabling SOA formation. The amount of SOA and its composition depends on the BVOC amount and mixture which are directly linked to the LU data sets. But the relationships between BVOCs, oxidants, and SOA concentrations are intricate, as they depend on the diurnal cycle of BVOC emission, transport, and degradation. Figure S23 in Supporting Information S4, shows the modeled total SOA concentrations. Maximum SOA concentrations are modeled for coniferous forests of the 10_LU and pine forests for the 138_LU. The monoterpenes grouped into the classes API donated α-pinene ( Figure S26 in Supporting Information S4) and LIM donated limonene ( Figure S27 in Supporting Information S4) contribute for almost the entire SOA difference between the two LU data sets. The fast night-time oxidation of API and LIM by NO 3 (in a few minutes Figure S43 in Supporting Information S7; Oderbolz et al., 2013) and its high SOA forming potential makes these SOA canals the most important once and is confirmed by its positive correlation ( Figures S24, S25  Monoterpenes are emitted throughout the day for the 10_LU. API and LIM concentration correlations are mainly positive ( Figure S24 in Supporting Information S4). The synthesis link of monoterpene emissions for spruce and beech combined with the rapid NO 3 reaction results in vast or even total night-time API and LIM concentration reductions ( Figures S50 and S46 in Supporting Information S7) which can induce negative correlations ( Figure  S25 in Supporting Information S4). 1,8-Cineol and p-cymene are characterized by an accumulation during night as their main oxidizing agent OH is then only scarcely produced, thus entail a positive correlation. However, direct SOA contributions are small. The higher 1,8-cineol emission of spruce and its link to diurnal syntheses cycle impairs correlation values. Otherwise BARO, BHC8, and OVOC concentrations seem to be extraordinary SOA proxies (Figures S45-S52 in Supporting Information S7). The night-time monoterpene emissions of pine trees facilitate SOA production via NO 3 oxidation turning pine into the major SOA contributor for the 138_LU. The difference between the two LU data sets is highest for spruce and beech followed by pine trees.
Isoprene ( Figure S28 in Supporting Information S4) and sesquiterpenes ( Figure S29 in Supporting Information S4) SOA concentrations are about one magnitude lower. Isoprene SOA concentration covers the entire land domain and is highest in the east and south-west of the domain for the 10_LU, while for the 138_LU it is only present in the south-west and far east. Brandenburg and the central model domain reveal the most significant isoprene SOA changes. Maximum sesquiterpene SOA concentrations are modeled in Brandenburg for both data sets. The huge SOA forming potential of sesquiterpenes is demonstrated by the difference plot ( Figure S29 in Supporting Information S4) as even small birch fractions yield considerable increases of SOA. As monoterpene NO 3 reaction products are the dominant SOA source, its night-time production is usually opposed to isoprene and sesquiterpene peak concentrations and thus negatively correlated. In areas with smaller negative OH SOA correlation (north and west of model domain) positive values are present.
The reduced BVOC emissions of the 138_LU (342 μg m −2 h −1 compared to 430 μg m −2 h −1 for 10_LU) diminishes the available absorbing OM which not only influences biogenic ( Figure S30 in Supporting Information S4) but also anthropogenic SOA ( Figure S31 in Supporting Information S4). Both are reduced with a minimum in the southern domain. Here, a decrease in overall SOA concentration by up to 60% is present, primarily caused by the reduced monoterpene emission of the 138_LU and the reduced OH availability due to the increased OH depletion by OVOCs which hinders SOA formation. Average anthropogenic SOA contribution is below 15%.
Considering SOA as part of PM 2.5 and PM 10 , changes in LU can affect PM composition. The simulated PM consists of primary emitted PM 10 and PM 2.5 , ammonium nitrate, ammonium sulfate, sulfate, sea salt, dust particles, and SOA. In general, in industrialized regions with strong anthropogenic emissions, such as the area between the Netherlands and the Ruhr area, the SOA fraction is typically less than 10%. In forests, SOA fraction values of up to 35% are reached for PM 2.5 ( Figure S32 in Supporting Information S4) and 27% for PM 10 ( Figure S33 in Supporting Information S4). Total difference between the two LU are about 6% for PM 2.5 and 5% for PM 10 with largest deviation between coniferous forests and spruce as well as pine trees.

Role of Additional Agricultural NO Soil Emissions
Changes to the oxidant concentrations by modified NO x regimes invoke LU depending air quality changes due to the specific BVOC mixtures as well as emission strengths and their response to NO x variations. Therefore, the whole setup is used to further investigate the interplay between LU induced changes to BVOC composition and strength combined with additional NO soil emission on air quality. The BVOC emission stays as previously stated for both LU data sets.
To test the NO model sensitivity, simulations with additional agricultural NO soil emissions have been performed for both LU data sets using the parametrizations from Williams et al. (1992) and Stohl et al. (1996). The agricultural NO soil parameterization is based on empirical relationships between soil temperature and land use type including fertilization rate of cropland. All plots already referred to in the previous sections show the additional NO emission scenario entitled with-NO on the right-hand side. The general patterns of the plots stay roughly the same. NO emission ( Figure S34 in Supporting Information S5) increased (57%) especially in the NW-SE agricultural band (Figure 1 and Figure S1 in Supporting Information S1). NO emissions and concentrations as well as NO 2 concentration patterns ( Figures S21 and S22 in Supporting Information S3) are comparable for both LU data sets. An increase in mean NO (40%), NO 2 (43%), and SOA (1%) and a decrease in ozone (5%) concentration was induced.
Additional NO soil emission influences the photostationary state between ozone, NO, and NO 2 (R1-R6) . This leads to ozone reduction and NO 2 rise independent of the land use, which are congruent to the NO emission increase ( Figure S34 in Supporting Information S5). In already high NO x regimes, for example, big cities followed by cropland NO concentration rises most notably ( Figure S35 in Supporting Information S5). Between the two LU data sets the lower NO and NO 2 concentrations in the western model domain, the Czech Republic and northern Poland of Figures S21 and S22 in Supporting Information S3 relate to the LU category natural and agricultural vegetation (Figure 1c) for which BVOC and NO emissions are slightly lower compared to the generalized mixed LU class of 10_LU.
Areas of low BVOC emissions are characterized by OH concentration reduction ( Figure S35 in Supporting Information S5). This applies for cropland, mud flats, pastures, wetlands, cities, and motorways. Here, the oxidation of O 3 by NO reduces the ozone concentration (R5) which hinders ozone photolysis (R1a) and OH production (R3). Furthermore, the lower BVOC concentrations cannot compensate the OH radical loss via OH production (R8 and R9). The production of NO 3 is linked to the reaction of NO 2 with ozone (R10) or the reaction of HNO 3 with OH (Stockwell et al., 1997). Here the reduction in ozone and OH concentration results in NO 3 loss ( Figure  S35 in Supporting Information S5) reducing the overall oxidant concentration in low BVOC concentration areas.
In areas of intensified BVOC emissions additional NO molecules facilitate the reaction between RO 2 and NO (R8) which enhances not only the NO 2 and HO 2 (Figures S34 and S35 in Supporting Information S5) but also the ozone production (R1 till R9). HO 2 declines with rising BVOC and declining NO x concentrations generally in the east and center of the model domain ( Figure S35 in Supporting Information S5). The formed OH radicals (R9) can initiate additional BVOC oxidation generating a self-reinforcing effect. Even though ozone concentration declines with increased NO emission for both LU data sets, the difference between the two widens ( Figure S18 in Supporting Information S3). Here, the higher isoprene and limonene emission of conifer forests in Brandenburg for the 10_LU and the isoprene oak emission spots in the west for the 138_LU trigger ozone production. The boosted OH concentration in BVOC rich environments through R7, R8, and R9 also enhances HNO 3 production (R13) with largest increase in the eastern model domain ( Figure S35 in Supporting Information S5). Here, the higher OH, HNO 3 , and ozone concentrations result in boosted NO 3 concentrations ( Figure S35 in Supporting Information S5) increasing the overall oxidant concentration in high BVOC concentration areas.
This increase in oxidants in BVOC-rich and low-NO x environments enhances BVOC degradation which lowers their concentration in coniferous forests for the 10_LU and for pine and oak for the 138_LU ( Figure S36 in Supporting Information S5). The larger BVOC emissions of the 10_LU creates a higher NO initiated boost. Under higher NO x conditions, the reduction of OH, NO 3 , and ozone hinders BVOC oxidation and leaves them untouched which increases BVOC concentrations mainly in the west and the NW-SE cropland band ( Figure S36 in Supporting Information S5).
As a result, isoprene and monoterpene SOA concentration rise in areas of lower NO x and high BVOC concentrations and diminish in the west with a minimum in the south of the Netherlands ( Figures S26-S28 in Supporting Information S4). Even though, Sesquiterpene oxidation is affected by the ozone reduction in BVOC-rich and high-NO x areas, a sesquiterpene-SOA increase is achieved in Brandenburg and Poland ( Figure S29 in Supporting Information S4). Here, the isoprene and monoterpene SOA growth increased the available OM intensifying SOA formation. The overall SOA concentration increases in BVOC rich environments and decreases for lower BVOC concentrations particularly close to the NO sources with a minimum in the south of the Netherlands ( Figure S36 in Supporting Information S5).

Impact of Agricultural Biomass Density Enhancements
Next to NO soil emission agriculture can impact BVOC emission strength by changes to the BD of the corresponding LU classes, for which different values are given in the literature (Karl, Guenther et al., 2009;Oderbolz et al., 2013;Steinbrecher et al., 2009). Therefore, a second sensitivity study was carried out investigating the influence of agricultural biomass enhancement. The BD of agro-forestry areas, pastures, non-irrigated land, permanently irrigated land, and agriculture mixed with natural vegetation was increased from 100, 40, 400, 400, and 50 g m −3 to 500, 250, 500, 500, and 375 g m −3 , respectively (main agricultural LU classes in the model domain N2 are: pastures, non-irrigated land, and agriculture mixed with natural vegetation). As an orientation values from Oderbolz et al. (2013) for agriculture (500 g m −3 ) and grassland (250 g m −3 ) are used. All LU classes have the same BVOC SEPs, therefore the increase in BVOC emissions is in line with the BD rise. Largest intensification can be found for pastures and mixed agricultural natural vegetation followed by non-irrigated cropland (Figure 1 and Figure S37 in Supporting Information S5). OVOC emission increased the most followed by α-pinene, limonene, and isoprene. No changes occur for sesquiterpenes, as agriculture was defined as non-sesquiterpene emitting (see Section 2.3). The additional BVOCs reduce OH (3.5%) and NO 3 (10%) concentrations while ozone increases by 0.5% ( Figure S37 in Supporting Information S5). For ozone, the production mechanism (R7 followed by R8, R9, and R4) compensates the consumption by BVOCs. This BVOC-triggered ozone increase reduces NO x concentrations especially in already highly NO x loaded environments (R5 till R10) and boosts BVOC oxidation resulting in enhanced SOA concentrations of up to 8% ( Figure S37 in Supporting Information S5).
Not only tree species but also different crop species are likely to impact air quality. So far, the model domain contains information about non-irrigated arable land and complex cultivated pattern but no crop speciation which can impact BD and subsequent BVOC emission composition and strength. Therefore, more research is needed to further investigate crop induced air quality changes.

Impact of Spatial and Temporal Model Resolution
For better comparability with measurements and clearer identification of regional effects, a simulation with finer resolution was performed for a smaller model domain (N3) covering eastern Germany including the TROPOS filed site Melpitz ( Figure S1 in Supporting Information S1). A nested grid starting from a resolution of 2.8 km, 1.4 km, and reaching 700 m for an area around Leipzig was used. Temporal resolution will adapt accordingly to the implemented IMEX scheme (see Section 2.1). The resolution can impact the meteorological parameters which itself control BVOC emissions and dispersion. The setup with the enhanced BDs for agricultural LU classes is used. The results of the simulation with finer resolution are shown in Figure S38 in Supporting Information S5.
Isoprene emission spots occur in the riparian forests SW and NW of Melpitz which are again linked to lower OH concentrations. α-Pinene emission is most intense for Brandenburg's pine forests reducing NO 3 concentrations. One limonene, 1,8-cineol, and p-cymene emission spot is present NW of Melpitz which can be matched to Pinus nigra. Birch trees NW and south of Melpitz lead to enhanced sesquiterpene emissions followed by pine and spruce trees. OVOC emission is highest in the southern domain for spruce trees. The resolution now also allows to draw conclusions on the influence of high NO x emitting roads (motorways, main roads) and demonstrates the importance of higher resolution simulations for regional air quality assessments. Motorways can clearly be identified in the OH concentration plot by the darker red and the lighter blue colors of the HO 2 and NO 3 plot. In mixed regimes, where motorways are located close to areas with higher BVOC emission (mainly Brandenburg) ozone and OH radicals are produced (R4 till R9). But directly along motorways, a reduction of ozone and NO 3 radical concentrations (R4-R6 and R10) occurs. Highest ozone concentrations are reached in south Brandenburg and in the south-east corner of the model domain where motorways are close but do not encounter. A possible link to ozone might also be valid for limonene emissions, but due to lacking sources no conclusion can be drawn. As isoprene emissions mainly collides with NO x emission sources, the effect of isoprene induced ozone production overlaps with ozone depletion by NO x . SOA concentration is highest for Brandenburg's pine trees.
Finer resolution enables a closer look at feedback mechanisms between NO x , BVOC, and LU. High NO x sources mainly motorways are clearly visible now and the detailed LU allows conclusions regarding curtain LU classes and their air quality impacts. There is evidence, that the small Pinus nigra forest close to the motorway with its high limonene emissions and which degradation is boosted by the increased OH concentration at motorways facilitates ozone and SOA production. These local effects are not visible in the coarser resolution of the N2 simulations. Higher resolution is therefore crucial when looking at feedback mechanisms between biosphere and troposphere and consequently for regional air quality assessments. All simulations including the high-resolution run are used for comparisons with field measurements (see below).

Comparisons With Measurements From Melpitz
The output of the different model scenarios were compared to measured NO, NO 2 (Thermo-Scientific 42i Meteorological parameters temperature, relative humidity, wind speed, and wind direction are measured at 6 m, global radiation at 2 m, and precipitation at 1 m height. The simulation output is at 10 m height. The simulated wind speed was corrected using a roughness length of 0.03 m, the most likely value for natural grassland and pastures (Silva et al., 2007). Tables S5 and S6 in Supporting Information S1 show statistical analysis (min., mean., max, standard deviation, correlation R, and slope) for the time period May 2014 and Figure 4 and Figures  S39, S40, S41 in Supporting Information S6 show the time series for OM concentration, meteorological parameters, gas-phase concentrations and SOA composition, respectively. Figure S45 in Supporting Information S7 shows the normalized time series and correlation matrixes for both LU data sets. Correlations between the measurements and the different simulations improve with resolution. Modeled wind speed is slightly higher (mean of 4.6 m s −1 for N1, 4.5 m s −1 for N2, and 4.2 ms −1 for N3) with an approximate offset of about 1 m s −1 and a stronger north-west wind component ( Figure S39 in Supporting Information S6) compared to the measurements (mean 2.9 m s −1 ). Temperature variation is similar for all simulations (R > 0.9). N1 shows an over as well as underestimation of temperatures on certain days ( Figure S39 in Supporting Information S6) while N2 and N3 capture the diurnal cycle well, but due to overestimated night temperatures (up to 3°C) higher mean temperature values are reached.
Gas-phase concentration of NO, NO 2 , and ozone ( Figure S40 in Supporting Information S6) as well as OM ( Figure 4) correlate best for higher BVOC and additional NO emission of the N2_10_NO simulation (Table S5 in Supporting Information S1). But due to the additional NO emissions, the mean values of NO and NO 2 for N2_10_ NO and N2_138_NO are highly overestimated, as they more than double. This consequently reduces ozone mean concentration by 6 μg m −3 (8%). During the entire time period, maximum differences of up to 31 μg m −3 (39%) occur. The gas-phase concentrations are clearly affected by NO emission but the two different LU data sets have only a minor impact for Melpitz as the dominant LU is agriculture, which is alike for both data sets.
The extension of the original SORGAM mechanism significantly improves OM concentration simulations and enhance the average value from 3.1 to 5.8 μg m −3 for the 10_LU_NO. During the first half of May 2014 the simulations are similar and reveal a strong anthropogenic SOA contribution (Figure 4 and Figure S41 in Supporting Information S6) of up to 70% for the original SORGAM and 50% for the extended SORGAM mechanism. Clear differences between the two SOA modules are evident for the second half of the month. The original mechanism shows moderated diurnal cycles, while for the extension clear peaks (20.-23.05, 26.-27.05.2014) are visible, which are in better agreement with the measurements. OM peak concentrations occur during night. N3 simulation captures the peaks best especially at the 26.05.2014. The anthropogenic portion of OM shrinks, reaching between 10% and 55% for the original and 5%-35% for the extended SORGAM mechanism. The main biogenic SOA sources of the original SORGAM are α-pinene + ozone, followed by limonene + ozone, α-pinene + OH, and limonene + OH reaction products. Other biogenic SOA sources are not included. The extended version reveals α-pinene + NO 3 SOA products as the dominating SOA source at the field site Melpitz reaching values between 10% and 65% of the total organic mass with a maximum during night. A closer look at the meteorological conditions ( Figure S39 in Supporting Information S6) during OM peak events reveal low wind speeds (below 0.5 m s −1 ), high temperatures and intense solar radiation during the day. This triggers BVOC emissions and reduces their spatial distribution. Therefore, they stay more local and due to progressing chemical degradation reaction products accumulate (SOA, O 3 , and NO 2 ). Correlation between AMS measurements and the simulation improves from R = 0.3 for the original to about 0.7 for simulations with the extended SORGAM version. Note that AMS measures OM concentration of PM 1 while the simulations predict OM of PM 10 , therefore higher simulated values are anticipated.

Summary and Conclusions
To examine the interaction between biosphere and atmosphere, first a tree species-specific BVOC emission split was implemented into the model system COSMO-MUSCAT and integrated into the chemistry mechanism RACM-MIM2ext and secondly applied in detailed gas-phase model simulations. Furthermore, the SOA module SORGAM was further extended. It now includes more BVOC reaction products and SOA precursors, and even provides detailed information about both the BVOC precursors and the oxidizing agents. In this way, the importance of the individual components, oxidants, and reaction pathways can be analyzed. For the simulations, a detailed LU data set containing more than 100 tree species was compared with a LU data set including only common LU categories, such as conifer and mixed forests. Additional agricultural sensitivity studies were carried out to investigate both the impact of NO soil emission and BD change on the interplay between NO x and BVOC emission as well as ozone, OH, NO 3, and SOA concentration, respectively.
The results from the comparison of the two different LU data sets demonstrates impressively that the detail of the implemented LU data set considerably affects BVOC emission strengths, composition, as well as emission patterns. Their degradation processes impact the chemical composition of the troposphere including air pollution, such as NO x , ground level ozone and PM (in terms of SOA). Here, the generalized LU implementation 10_LU in the case of Germany leads to 50% overestimated BVOC (excluding OVOCs) emissions generating altered oxidant concentrations (NO x ±2.5%, ozone +2.5%, OH ±50%, and NO 3 radical −70%) and a 60% overestimation in SOA concentration compared to detailed tree species information. Jiang et al. (2019) also found only minor impacts on ozone but significant changes in SOA concentrations when comparing two different biogenic emission models for Europe.
BVOC emission spots are evident where tree species emit higher amounts of BVOCs than their corresponding forest category: (a) oaks for isoprene, (b) birch and spruce for sesquiterpenes, (c) beech and spruce for OVOCs, (d) beech for API (mainly sabinene), and (e) spruce for 1,8-cineol. Every single BVOC has its preferred oxidizing partner. Therefore, BVOC degradation strongly effects the distribution and concentration of their oxidants, such as ozone, OH, and NO 3 radicals. The present study revealed that isoprene concentration is inversely correlated to OH and monoterpene to NO 3 radical concentrations. The lack of isoprene emission of pine trees induces an increase in OH while oaks have the utmost isoprene emission strength decreasing OH radical concentration and generating ozone during chemical conversion. The second major OH sinks are OVOCs and its degradation hinders SOA production diminishing SOA concentration in spruce forests compared to their generalized LU class coniferous forest. Even though OVOCs do not contribute to SOA formation, they are an excellent proxy for temporal SOA evolvement. Monoterpenes are emitted in huge amounts by conifers and in the case of α-pinene also by beech. Monoterpene NO 3 reaction products are the major SOA components in the model domain. Therefore, coniferous forests of the 10_LU and pine of the 138_LU are the major SOA sources but a 30% reduction of SOA manifests for pine forests mainly due to lacking limonene SOA products. LU controls the parameters defining the BVOC emission strengths and composition thereby regulating oxidant concentrations while their chemical degradation path defines their ozone and SOA forming potential. BVOCs are a crucial link to air quality and climate as this will not only impact radiation but also cloud and precipitation formation.
Sensitivity studies regarding agricultural BD enhancement reveals a regional increase in BVOC emissions (50%), ozone (0.5%), and SOA (8%) while OH (3.5%) and NO 3 (10%) concentrations decline. Additional agricultural NO soil emissions (57%) induce an equivalent NO 2 concentration (42%) increase but on the other hand an ozone (5%) and NO 3 (20%) radical concentration reduction. High NO x environments show most significant NO increases and in combination with low BVOC concentrations OH radical concentration declines. Here the overall reduction of oxidizing agents inhibits BVOC degradation and consequently lowers SOA concentrations. In BVOC dominated lower NO x areas, the BVOC degradation under increased NO emissions leads to the production of OH radicals and ozone. However, the production of ozone coincides with ozone depletion by NO. The additional ozone and OH radicals enhance BVOC oxidation which facilitates SOA production (+1%). The higher BVOC emission strength of the 10_LU leads to intensified ozone and SOA production compared to the 138_LU which further enhances the differences between the two data sets while at the same time OH and NO 3 concentration differences decrease in BVOC dominated areas. Here, we focused on tree species-specific impacts on air quality but also crop induced air quality changes have to be considered for air quality assessments. But, more knowledge is needed on crop specific parameters and parameterizations.
Comparisons with measurements from Melpitz reveal a significant improvement of simulated organic matter concentrations (from R of 0.3 to 0.7) when extending the SOA module. Monoterpene products are identified as the main SOA components (up to 90%), solely α-pinene + NO 3 SOA products make up to 65%. Low wind velocities, warm temperatures and high solar radiation can facilitate BVOC-induced air quality impairment which makes these cases even more sensitive to the underlying LU. Slightly changed wind conditions can alter the dispersion of VOCs as well as their oxidizing agents, and temperature can influence not only the emission of BVOCs but also their reaction speed and their partitioning into the particle phase. An overestimation of the wind speed and night time temperature leads partly to an underestimation of SOA concentration. The exact capture of meteorological conditions is therefore, another key point. Meteorological performance benefits from higher spatial and temporal resolution. Here, an improvement was achieved with every refinement from N1 to N2 and finally N3. High resolution simulations are crucial for analyzing local effects on the BVOC degradation in clean as well as in NO x -influenced environments and enables a more detailed investigation into feedback mechanisms between NO x , BVOC, and LU.
Overall, the simulations showed that BVOCs play a major role in air quality assessment which should be addressed by a diverse SEP, BD, and a tree species-specific emission split. Unfortunately, BVOC emission strengths as well as their exact composition remain uncertain. BVOC emissions are strongly impacted by the emission parameterization, the driving meteorological parameters and LU. For air quality assessments, also the applied chemistry mechanism and SOA module are of key importance. To further address this complex issue and to include more species-specific dependencies linked to BVOC emission parametrization as well as meteorological effects more research is needed. Here, we followed the recommendations from Steinbrecher et al. (2009) and managed to obtain a flexible model setup which allows targeted tree species-specific BVOC emissions directly integrated and online coupled to guarantee immediate biosphere-atmosphere interactions. While we incorporated the BVOC emission scheme, we identified further potential species-specific integration possibilities for several aspects regarding LU, BVOC emission parameterization, meteorology, chemistry, and SOA treatment which, to our knowledge, have not been taken into account so far and are discussed in more detail in the Supporting Information S8 (see Text S1).

Outlook
This work demonstrates the high sensitivity of BVOC emissions toward LU cover implying huge potential for further LU database improvements when considering tree species-specific data. Furthermore, an advanced knowledge about the individual species-specific response to a variety of stressors, such as ozone, drought, wounding, CO 2 , etc. is important to further improve BVOC emission estimates as this will modify BVOC emission strengths and composition and thus affect air quality through ozone and SOA formation. This impacts not only radiation but also cloud and precipitation formation processes and thus entails a feedback mechanism back to climate. Therefore, more knowledge is needed about the biosphere-atmosphere interaction in view of a changing climate. But so far, only a few studies on stress response exist. The findings of the present air quality study, aimed to study the impacts of a forest versus detailed tree species LU data set, are linked not only to LU data sets but to the chemical mechanism and SOA treatment as well. To further improve the chemical degradation of biogenic and anthropogenic VOCs, the currently applied gas-phase chemistry mechanism RACM-MIM2-ext will soon be replaced by a more advanced chemical mechanism which is based on the MOZART mechanism (Schultz et al., 2018). This further improvement will help to overcome the discrepancies between the measured and simulated gas-phase concentrations. Simultaneously, the SOA module has to be adjusted and extended in accordance with the new gas-phase mechanism. These developments shall further improve air quality assessments on a regional to global scale. The feedback between transport processes, biosphere and atmosphere will be further investigated focusing on the interaction of biogenic and anthropogenic emissions and their influence on air quality (PM, O 3 , and NO x ). For better comparisons with measurements, a size-resolved SOA model treatment would considerably benefit further analyses.