The influence of weather-driven processes on tropospheric ozone

Near-surface ozone is an harmful air pollutant, which is determined to a considerable extent by weather-controlled processes, and may be significantly impacted by water vapour forming complexes with peroxy radicals. The role of water in the reaction of HO2 radical with nitrogen oxides is known from the literature, and in current models the water complex is considered by assuming a linear dependence on water concentrations. In fact, recent experimental evidence has been published, showing the significant role of water on the kinetics of one of the most important reaction for ozone chemistry, namely 5 NO2 + OH. Here, the available kinetic data for the HOx + NOx reactions have been included in the atmospheric chemistry model ECHAM5/MESSy (EMAC) to test its global significance. Among the modified kinetics, the newly added HNO3 channel from HO2 + NO, dominates, significantly reducing NO2. A major removal process of near-surface ozone is dry deposition accounting for 20 % of the total tropospheric ozone loss mostly occurring over vegetation. However, parameterizations for modelling dry deposition represent a major source of uncertainty for tropospheric ozone simulations. This potentially be10 longs to the reasons why global models, such as EMAC used here, overestimate ozone with respect to observations. In fact, the employed parameterization is hardly sensitive to local meteorological conditions (e.g., humidity) and lacks non-stomatal deposition. In this study, a dry deposition scheme including these features have been used in EMAC, affecting not only the deposition of ozone but of its precursors, resulting in lower chemical production of ozone. Additionally, we improved the emissions of isoprene and nitrous acid (HONO). Namely, for isoprene emissions we have accounted for the impact of drought 15 stress which confers a higher model sensitivity to meteorology leading to reduced annual emissions down to 32 %. For HONO, we have implemented soil emissions, which depend on soil moisture and thus on precipitation. We estimate for the first time a global source strength of 7 Tg(N) a−1. Furthermore, the usage of a parameterization for the production of lightning NOx that depends on cloud top height contributes to a more realistic representation of NO2 columns over remote oceans with respect to the satellite measurements of the Ozone Monitoring Instrument (OMI). The combination of all the model modifications reduces 20 the simulated global ozone burden by ≈ 20 % to 337 Tg, which is in better agreement with recent estimates. By comparing simulation results with measurements from the Infrared Atmospheric Sounding Interferometer (IASI) and the Tropospheric Ozone Assessment Report (TOAR) databases (of 2009) we demonstrate an overall reduction of the ozone bias by a factor of 2. 1 https://doi.org/10.5194/acp-2021-584 Preprint. Discussion started: 16 July 2021 c © Author(s) 2021. CC BY 4.0 License.


25
The investigation of tropospheric ozone (O 3 ) plays a prominent role in atmospheric research. In fact, O 3 is an air pollutant harmful for humans and plants, has a significant radiative forcing, and is an important oxidant in the troposphere Fleming et al., 2018). Its levels are strongly affected by radical reactions and surface-atmosphere exchanges which in turn are modulated, directly and indirectly, by weather (Fiore et al., 2012;Jacob and Winner, 2009;Sadiq et al., 2017;Fu and Tian, 2019). Understanding the impact of weather on ozone, and air quality in general, is thus important also in view of 30 weather extremes. In fact, the frequency and intensity of heat waves and droughts are projected to increase due to climate change (Hou and Wu, 2016). Global chemistry-climate models (CCM) overestimate tropospheric ozone by about 20 % in the Northern Hemisphere with respect to observations (Young et al., 2018(Young et al., , 2013. The global atmospheric chemistry model ECHAM/MESSy (EMAC; Jöckel et al., 2010) is no exception to this since Jöckel et al. (2016) reported that EMAC tends to overpredict tropospheric O 3 columns by up to 15 DU 1 (see their Fig. 29) when compared to satellite retrievals. Recently, 35 by using the Jülich Aqueous-phase Mechanism of Organic Chemistry (JAMOC; Rosanka et al., 2021a) in EMAC, Rosanka et al. (2021b) identified that the simplified representation of in-cloud organic chemistry may account for up to 20 % of this model bias. In addition, Rosanka et al. (2020) reported that a comprehensive representation of this process is important for estimating the impact of intense peat fires on tropospheric O 3 . Still, an ongoing model development and assessment with a focus tropospheric ozone is needed to further reduce the high model bias. The meteorological dependence of processes driving 40 ozone has been well reported in the literature (Kavassalis and Murphy, 2017, and references therein). The ozone levels near the surface are determined by chemical production and loss reactions involving nitrogen oxides (NO x =NO+NO 2 ) and volatile organic compounds (VOCs) (Monks, 2005). Radical reactions are known to be affected in the presence of water vapour. Water vapour is the third most abundant species in the troposphere with the largest abundance at the lowest levels. The ability to form stable complexes with atmospheric radicals is acknowledged which modifies the kinetics of the HO x (HO x =OH+HO 2 ) and 45 NO x reactions which are key to photo-chemical ozone production. However, the relevant kinetics is not entirely known and included only partially and in a simplified way in models (Buszek et al., 2011). Dry deposition represents an important sink accounting for 20 % of the total O 3 loss in the troposphere (Young et al., 2018) which is the highest over vegetation (Hardacre et al., 2015). However, the representation of dry deposition in models is uncertain due to the limited amount of observations and the dependence on input data (Val Martin et al., 2014;Wong et al., 2019;Wesely and Hicks, 2000). A potential uncertainty 50 source is the parameterization of surface resistance as identified in a multi-model evaluation by Hardacre et al. (2015). Schwede et al. (2011) points to soil and cuticular (wax coverage of leaves) uptake. Also, the dependence on meteorology plays a major role for the prediction of dry deposition and its implementation becomes especially desirable in the light of a changing climate (Andersson and Engardt, 2010;Wong et al., 2019). A further important process affecting the chemistry and fate of tropospheric ozone is the biogenic emission of isoprene (C 5 H 8 ), the most important volatile organic compound (BVOC) emitted from plants The relevant atmospheric gas-phase chemistry in the troposphere and stratosphere is calculated within the submodel MECCA (Module Efficiently Calculating the Chemistry of the Atmosphere) using the Mainz Organic Mechanism (MOM; Sander et al., 2019), which contains an extensive oxidation scheme for isoprene (Taraborrelli et al., 2009(Taraborrelli et al., , 2012Nölscher et al., 2014), monoterpenes (Hens et al., 2014), and aromatics (Cabrera-Perez et al., 2016;Taraborrelli et al., 2021) excluding iodine and mercury chemistry. The resulting system of ordinary differential equations (ODEs) represents 697 gaseous species and 2099 100 reactions. MECCA employs a sophisticated tagging system that allows for obtaining reaction rates from multiple reactions and combining them into a single tracer (Gromov et al., 2010). This system allows to obtain detailed tropospheric budgets of tracers. When considering the global O 3 budget, odd oxygen (O x ) is analysed to account for rapid cycling between species of the O x family. In the scope of this work, O x is defined as: O x ≡ O + O 3 + NO 2 + 2 × NO 3 + 3 × N 2 O 5 + HNO 3 + HNO 4 + ClO + HOCl + ClNO 2 + 2 × ClNO 3 + BrO + HOBr + BrNO 2 + 2 × BrNO 3 + PANs + PNs + ANs + NPs (1) 105 where PANs are peroxyacyl nitrates, PNs are alkyl peroxynitrates, ANs are alkyl nitrates, and NPs are nitrophenols. For the tropopsheric O x budget presented in Table 3, the tropopause is defined in the extratropics using potential vorticity, whereas temperature lapse rates are used in the tropics (Jöckel et al., 2010).
The removal of trace gases and aerosol particles by clouds and precipitation is simulated by the SCAVenging submodel (SCAV; Tost et al., 2006). SCAV calculates the transfer of species into and out of rain and cloud droplets using the Henry's 110 law equilibrium, acid dissociation equilibria, oxidation-reduction reactions, heterogeneous reactions on droplet surfaces, and aqueous-phase photolysis reactions representing more than 150 reactions (Tost et al., 2007a).
The submodel DDEP provides the dry deposition of trace gases  in which the dry deposition at the surface is determined based on the standard resistance-in-series model by Wesely (1989). Biogenic emissions of VOCs are parameterised with an emission activity algorithm by the submodel MEGAN (Guenther et al., 2006) and soil emissions of NO 115 are calculated according to an empirical model by Yienger (1995) in the submodel ONEMIS. Anthropogenic emissions are based on data from the RCP8.5 scenario performed for the fifth assessment report (AR5) of the Intergovernmental Panel on Climate Change (IPCC) (Lamarque et al., 2010), vertically distributed following Pozzer et al. (2009). For representing biomass burning emissions the Global Fire Assimilation System (GFAS) inventory is used which contains observed dry matter burn and factor of ten lower production for intra-cloud (IC) flashes and the flash frequency is scaled to 6.548 as in Jöckel et al. (2016).
This yields an annual global production efficiency of 318 mol(NO) flash −1 which agrees with the reported estimates of 250 -400 mol(NO) flash −1 of known studies (Gordillo-Vázquez et al., 2019;Miyazaki et al., 2014;Schumann and Huntrieser, 2007 each including an analysis of the global changes. The respective budgets of O x are given in Table 3. 3 Observational data 3.1 Station measurements from the Tropospheric Ozone Assessment report (TOAR)

135
For evaluating surface ozone, we use ground-based station measurements collected in the TOAR database . The database comprises in-situ hourly data from almost 10,000 measurement sides around the globe which are available for the period 1970-2015. These have been collected from different ozone monitoring networks (e.g., Clean Air Status and Trends Network (CASTNET)) and data providers (e.g., Umweltbundesamt), harmonised and synthesised to enable an uniform processing. The data were selected based on an extended quality control; e.g., sites where the measurement technique changed 140 with time have been excluded. Data errors still remain but have been shown to have a minor impact . The total uncertainty in modern ozone measurements is estimated with < 2 nmol mol −1 (Tarasick et al., 2019). The here used O 3 product is shown in Figure 1.

O 3 IASI satellite product
The satellite measurements of tropospheric ozone used here to evaluate the EMAC simulations are based on the observations by 145 the Infrared Atmospheric Sounding Interferometer (IASI) instrument, a thermal infrared (TIR) Fourier transform spectrometer which measures the backwards radiation of the Earth's surface and of the lower atmosphere in a nadir-viewing geometry (Clerbaux et al., 2009) (Hurtmans et al., 2012). The quality of the IASI profile product is ensured by the application of a series of specific quality flags rejecting poor spectral fits, data with poor vertical sensitivity and cloud contaminated IASI scenes . The IASI tropospheric column of ozone is defined as ranging between the ground and 300 hPa. This range also allows limiting the influence of stratospheric O 3 on the retrieved tropospheric column, while still including the layers 155 of maximum sensitivity of IASI in the troposphere (Wespes et al., 2017). The total statistical error from the retrievals for the tropopsheric column is estimated to to about 5-20 % (Hurtmans et al., 2012;Wespes et al., 2016;Boynard et al., 2018). For the here used partial column up to 300 hPa, Boynard et al. (2018) have reported a negative bias by O 3 -FORLI in the mid-latitudes and in the tropics (11-13 % and 16-19 %, respectively) compared to ozonesonde data. In order to account for the vertical sensitivity of IASI to O 3 measurements, the averaging kernels associated with each retrieved O 3 profiles were considered for 160 the model-to-satellite comparisons. In this context, we used the MESSy submodel SORBIT (Jöckel et al., 2010) to sample the complete EMAC vertical profiles (in volume mixing ratio; VMR) at the time and location of the IASI measurements.
The co-located EMAC profile was first interpolated to the FORLI-O3 pressure grids and then converted into column profile.
Then, the altitude-dependent sensitivity of IASI has been taken into account by applying the FORLI-O3 averaging kernels on the co-located EMAC profile following the formalism of Rodgers (2000). The tropospheric O 3 columns were then calculated 165 from the convoluted profiles between the ground and 300 hPa. Since several IASI observations usually fall into a given model 2.8 • × 2.8 • grid box for a given day, the interpolation/convolution of the corresponding EMAC profile was repeated separately for each IASI measurements falling in that model grid box on that day (see, e.g., Schultz et al., 2018;Rosanka et al., 2021b). A daily mean value of O 3 tropospheric column was eventually calculated for that grid box by averaging all the columns obtained from the repeated smoothing of the EMAC profile. The application of the IASI averaging kernels to the EMAC ozone profiles, 170 returns simulated tropopsheric O3 as seen by IASI. The resulting product is shown in Figure 2.

NO 2 OMI satellite product
Due to the potential impact of the model developments on the tropospheric level of NO 2 which is an important precursor of tropospheric ozone, we use the NO 2 tropospheric columns measured by the Ozone Monitoring Instrument (OMI). OMI has been launched on board NASA's Earth Observing System Aura satellite in July 2004. The nadir-viewing near-UV-visible 175 (400-470 nm) spectrometer measures the Earth's backscattered radiance and solar radiance in a Sun-synchronous polar orbit, crossing the equator at 13:40 local solar time (Zara et al., 2018). From this, the total slant density column (SCD) of a trace gas (concentration integrated along the light path between the earth surface and the satellite) is calculated using the differential optical absorption spectroscopy (DOAS) method based on the Beer-Lambert law. The stratospheric part of the SCD is estimated by the Data assimilation transport model TM5 (http://www.qa4ecv.eu/ecv/no2/main/strato) which is then subtracted from the 180 total SCD to retrieve the tropospheric SCD. Dividing by the tropospheric air mass factors (AMFs) yields the tropospheric vertical column density (VCD). For a harmonised comparison to account for OMI sensitivity, the averaging kernel provided by the QA4ECV product has been applied whereas the original (TM5) a priori profiles from OMI QA4ECV product have been replaced with that from the EMAC simulation. The NO 2 VCD product used in this study has been developed by the Quality yields the smallest uncertainties (0.8 × 10 15 molecules cm −2 ) comparing the previously existing NO 2 products. In general, surface and atmospheric properties (e.g. cloudiness, solar zenith angle, snow cover, large aerosol close to the surface) largely affect the vertical sensitivity of the instrument. Therefore, the data used here was filtered with discarding those corresponding to processing error flag, cloud radiance fraction > 0.5, solar zenith angle > 80 • , snow covered land and ocean (> 10% snow cover) and the ratio of tropospheric airmass factor and geometric airmass factors < 0.2 (description of the flags in 190 the documentation). By that, however, cloud-free conditions are preferred for tropospheric VCDs and negative biases occur, especially over polluted regions. General uncertainties of the tropospheric NO 2 retrievals can be related to the separation of tropospheric from stratospheric SCD (Zara et al., 2018). Beirle et al. (2016) have shown that the separation of the stratospheric and tropospheric SCD in the calculation of the NO 2 satellite retrievals leads to a general uncertainty between 0.1 and 0.2 × 10 15 molecules cm −2 (5-10 %) which is negligible in high polluted areas but plays a role in moderately polluted areas. In comparison 195 with ground-based measurements acquired by MAX-DOAS the OMI QA4ECV products show a negative bias of 1-4 × 10 15 molecules cm −2 (Compernolle et al., 2020;Kumar et al., 2020).
In order to ensure the representativity during comparison with the simulations, data has been sampled along the satellite track by the MESSy submodel SORBIT (Jöckel et al., 2010). Then, this has been integrated over the location dependent tropopause level given in the QA4ECV product. The data is only analysed if at least 10 values exist. The resulting product is shown in 200 Figure 3.

The role of water vapour in radical reactions
Water vapour has a high abundance in the troposhere, in particular in the lowest layers where increasing levels are predicted during global warming. Besides its role as the most important greenhouse gas warming the Earth's atmosphere, water vapor is known to form complexes with radicals (Buszek et al., 2011). These complexes modify the reaction kinetics and often act as 205 catalyst enhancing the fate and the speed of the reactions (e.g., Zhang et al., 2019;Kumbhani et al., 2015;Buszek et al., 2011).
Since hydroxy and peroxy radicals are involved in the formation of photochemical species such as O 3 , the water-complexes influence the budgets of these trace gases as suggested by, e.g., Butkovskaya et al. (2009).

Modified kinetics of NO x
In MOM, the reaction rate constant of HO 2 + NO forming NO 2 and OH is represented originally according to Burkholder (2) In order to allow the formation of water complexes, simulated in EMACh2o, this is replaced by k 1w , which is defined as: where k 0 is the recommended value by Sander et al. (2003). Additionally to the NO 2 +OH channel, another HNO 3 formation channel, represented with the reaction rate constant k 2w , is considered explicitly for the first time in this study and added to EMACh2o. The importance of this formation channel has been clearly indicated by e.g., Righi et al. (2015).
where β, also used in Cariolle et al. (2008), is based on Butkovskaya et al. (2009) and is recommended by JPL (J. B. Burkholder and Wine, 2020). The detailed calculation of α is given in Butkovskaya et al. (2009); Gottschaldt et al. (2013). Following the 220 results of Duncianu et al. (2020), the 2-fold lower reaction rate k 2w is used here. The potential role of water complexes in the kinetics of organic peroxy radicals has been indicated in a few studies (Kumbhani et al., 2015;Clark et al., 2008Clark et al., , 2010. Recently, Xing et al. (2018) proposed that substituted RO 2 −water complexes during the oxidation of methacrolein might have a unity alkyl nitrate yield when reacting with NO and estimated a lower limit for the equilibrium constant for the complexation.
Although independent supporting evidence for the enhancement of the alkyl nitrate yield under humid conditions is warranted,

Impact on the tropospheric ozone chemistry
Allowing the formation of water complexes in radical reactions have significant impacts on the chemistry of O x as seen in Table 3. The chemical production and loss terms of O x in the troposphere are decreased by about 20 % compared to the reference simulation (EMACref), which brings the model in better agreement with the recent multi-model estimates of 4500-5200 Tg a −1 chemical production and 4000-4800 Tg a −1 loss (Young et al., 2018). As indicated by Butkovskaya et al.

235
(2009), who assumes that the HNO 3 channel at 50 % humidity is as important as the HNO 3 production from OH + NO 2 , the enhanced formation of HNO 3 from the reaction of HO 2 with NO including water complexes is dominant in many regions ( Fig. 4a). This reduces the NO 2 yield from the HO 2 + NO reaction as well as OH, and thus the oxidation capacity of the atmosphere (Gottschaldt et al., 2013). The additional produced HNO 3 , however, is deposited faster than converted into NO 2 and transported to the Southern oceans (decrease of NO 2 ) (Fig. 4b). Consequently, less O 3 is formed by NO 2 photolysis 240 which is strongest on the southern hemisphere continents. In higher polluted regions like North America, Europe and East Asia, the modified HO 2 +NO reaction affect significantly the HO 2 yield (Fig. 4c), also reported by Butkovskaya et al. (2009).
Due to the counter-balance with a slight increase of NO 2 the impact on O 3 is smaller than on the SH. The modified RO 2 +NO kinetics play a minor role as it can be estimated from the changed dry deposition flux of the produced alkyl nitrates ( AN;  Table 8). The global burden of the two ANs, i−C 3 H 7 NO 3 and n−C 3 H 7 NO 3 increases from 15.3 to 65.7 and 10.1 to 43.8 Gg, respectively, which is much higher than the estimates by Khan et al. (2015, Tab. 5). The higher burden in EMAC can be attributed to the rapid hydrolysis of isoprene nitrates (Vasquez et al., 2020) not being represented in the here used model simulation, which is known to reduce the burden and dry deposition of nitrates. From the dry deposition estimate a conclusion 250 for O 3 can be drawn since the RO 2 /NO reaction produce either one alkyl nitrate or two ozone molecules (major branch) (Day et al., 2003). In fact, the change of the AN dry deposition flux corresponds to 3.3 Tg a −1 less O 3 removal, which is minor relative to the absolute O 3 deposition (Tab. 3). The contribution of the RO 2 /NO reaction to the change of the O x production is with 7 % much smaller than the 70 % contribution by the HO 2 /NO reaction. Overall, the inclusion of water complexes decreases the mismatch between modelled and measured surface ozone by 10-15 % towards a small remaining bias of ± 4 255 nmol mol −1 (Fig. 11b,f). The highest mismatch occur in sub-urban coastal grid boxes where an accurate representation of chemical concentrations is notoriously challenging since the steep gradients of mixing ratios from land to ocean (stronger during daytime) are not resolved in models (Fiore et al., 2002). Also, the annual mean tropospheric column ozone is decreased by 10-20 % due to the modified kinetics ( Fig. 13b). Thus, the bias of EMAC relative to IASI is significantly reduced (between 40 • S and 40 • N), as shown in Figure 12a and Figure  where the used tropopause definition, is however different than the one used here. Also, Righi et al. (2015) have reported a significant reduction of the ozone bias in the troposphere due to the inclusion of the HNO 3 production from HO 2 +NO.
Contrarily, the model underestimation of tropospheric NO 2 in comparison to OMI is strengthened by EMACh2o (Fig. 16b).

The role of weather for dry deposition
Dry deposition of trace gases to vegetation depends on several meteorological variables which often determines the variability of ozone levels (Wong et al., 2019;Val Martin et al., 2014;Kavassalis and Murphy, 2017). In fact, the relation to air temperature is well known and has been reported by many studies (e.g., Jarvis, 1976;Hogg et al., 2007). The importance of linking ozone uptake with the atmospheric water demand has been shown by e.g. Fares et al. (2012); Hogg et al. (2007) to be essential 270 to capture the daily depression in the afternoon. In the light of global warming with higher temperatures and more frequent droughts, these links become increasingly important and the integration in models is desirable (Andersson and Engardt, 2010;Wong et al., 2019).

The advanced dry deposition scheme
The uptake of trace gases in EMAC is based on the 'big-leaf approach' and calculated by the means of multiple resistors which 275 represent the different spheres of the soil-vegetation-atmosphere system (e.g., Wesely and Hicks, 2000;Wesely, 1989). Both methods are commonly used in Earth system models due to its simplicity as reported in the review by Clifton et al. (2020) or as applied by e.g. Val Martin et al. (2014); Zhang et al. (2003). In this study, an extension of the general framework described by Kerkweg et al. (2006) is used in the EMACddep simulation. The extended scheme includes additional dependencies of the the uptake at the plants' cuticle (wax covering of the leaves) has been incorporated as well as the consistent usage of the Leaf Area Index (LAI), which is the basic vegetation information of the scheme, as described by Emmerichs et al. (2021).

Impact on the O x dry deposition and chemistry
The inclusion of the revised dry deposition scheme by Emmerichs et al. (2021) enables a proper cuticular uptake leading to a higher deposition over vegetated areas, which is amplified by the stomatal response in moderate climate (optimal temperature: 285 25 • C, humid air). Balancing effects occur in dry and hot climate as well as in the Tropics (highest LAI) due to the oversimplification of dry deposition in the reference scheme with a LAI of 1 (Emmerichs et al., 2021). Hence, the impact shows a spatial and temporal variation where O 3 dry deposition increases with the growing of vegetation on the northern hemispheric continents up to 20 %. In this area, surface ozone mixing ratios decrease by 5 to 10 %. If day-and nighttime values are studied separately the change by EMACddep plays relatively a more important role during nighttime over most vegetated areas as 290 shown in Figure 5 for boreal summer. Since the newly implemented cuticular uptake does not rely on sunlight, contrasting to the stomatal pathway. Therefore, also the balancing effects described above does not occur during nighttime. The revised dry deposition decreases the overestimation of EMAC with respect to the TOAR observation data during day and night by 5-10 %, where during the day the bias is generally lower with a remaining discrepancy of only up to ± 6 nmol mol −1 (Figs. 11c, g).  (Young et al., 2018). The representation of O 3 dry deposition could be further improved by explicitly representing O 3 deposition to soil, which several measurement studies report to be an important pathway (e.g. Fares et al., 2012;Stella et al., 2011). Moreover, we know that due to a dry bias (too dry soil and too high temperature) in the Amazon basin the dry deposition (Hagemann and Stacke, 2015) is highly underestimated in this area, 305 very important for dry deposition (Emmerichs et al., 2021).

Impact of the isoprene emission modelling
Isoprene is the most important biogenic emitted VOC accounting for 50-70 % of the total emissions (Guenther et al., 2006;Sindelarova et al., 2014). Common studies use the MEGAN model (Guenther et al., 2012) which estimates the emission fluxes with multiple activity factors accounting for different environmental responses. The here used MEGAN2.1 covers several of 310 these responses but lacks the representation of drought stress, since in conditions of less precipitation and dry soil (i.e. drought) the emissions of isoprene are affected where the response depends on the duration of the drought (Ferracci et al., 2020). Overall, a significant decline of isoprene emissions have been observed due to droughts (Pegoraro et al., 2004;Ferracci et al., 2020).
The representation of this response is expected to become more important in the light of global warming with an increasing number of droughts (Samaniego et al., 2018).

The global impact
Isoprene is mainly emitted in the tropical rain forests of Brazil, Central Africa and Indonesia (e.g. Sindelarova et al., 2014;Guenther et al., 2006;Müller et al., 2008;Henrot et al., 2017), such as simulated by EMAC (Fig. A3). The usage of the 325 drought factor (seasonal means in Fig. A2), which have been shown to improve the accuracy of the model in capturing drought and non-drought periods , leads to a decrease of the emission flux in the emission regions of about 20 mg m −2 s −1 in boreal summer (Fig. 6a). The major reaction partner OH, thus shows an increase which cause a reduction of O 3 at the surface (Fig. 6b). These results are similar to those presented by Jiang et al. (2018) applying the drought activity factor in CAMS-Chem. For the comparison to the observed tropospheric NO 2 column this effect is only important in relative terms.

330
Hence, also the difference of the EMAC and the IASI tropospheric ozone column is not influenced significantly. Globally, the annual isoprene emission flux is reduced from 508.2 to 347.8 Tg ( as the most comparable estimate  give no global estimate), we conclude overall that EMACisop represents reasonably the global isoprene emissions while it overcomes the major uncertainty related to the parameterization of soil 340 moisture stress for isoprene emissions . However, we note that the range of model estimates for the global isoprene emission is large due to the existing uncertainty in modelling (Arneth et al., 2011).
Nitrous acid (HONO) is an important OH source, especially in the morning, and thus it impacts significantly the HO x and O 3 budgets (Monks, 2005;Sörgel et al., 2011). However, current models tend to underestimate HONO levels, which is attributed 345 to an incomplete knowledge of HONO sources (e.g. Li et al., 2010). Recently, soil emissions have been shown to close this gap (Su et al., 2011;Yang et al., 2020) Figure   8a. The OH increase at the surface on the southern hemisphere comply with the reported negative bias of models (Naik et al., 2013). The inter-hemispheric disparity of OH (Lelieveld et al., 2016;Patra et al., 2014), however, is not influenced significantly 370 by EMAChono. The OH increase reduces the isoprene level at the surface over most regions. Since HONO also represents a source of NO x its increase enhances the annual mean tropospheric NO 2 column by up to 0.5 × 10 15 molecules cm −2 (60 %) which is the highest over Australia and Central Africa. But this has a minor impact on the EMAC/OMI bias. The enhanced NO 2 and HONO levels add up to a small relative increase of tropospheric ozone (up to 6 %) in the Tropics. All these changes influence the global chemical production and loss of O x . In the planetary boundary layer (PBL), the O x chemical production and loss is increased by 12 % and 6 %, respectively, due to the modification which yields a net increase of O x by chemistry (details in Tab. 3. This increase is also shown in Figure 8 for ground-level ozone over the SH continents (no observations available). This might improve the known underestimation of ground-level ozone in the Southern Hemisphere (Young et al., 2013). Regarding the whole troposphere, also a net increase is seen, but this change is minor ( 4.5 %).
Overall, this development represents one of the first implementation of an additional HONO source, the emission from soil

The lightning NO x parameterization
The LNO x scheme by Grewe et al. (2001), used in EMACref as the current standard in EMAC, overpredicts lightning over the ocean since the flash activity is treated equally over land and oceans. Furthermore, the data assimilation of multiple satellite 395 data by Miyazaki et al. (2014) suggests that the C-shape based parameterizations, such as the Grewe scheme, overestimate the peak source height by up to 1 km over land and the tropical western Pacific. In contrast, the scheme by Price and Rind (1992) (P&R), which is commonly used e.g. for the CMIP6 simulations (Griffiths et al., 2021), stands out by representing lightning over land and ocean with distinct flash frequencies. Also, it is described as robust in space and time. Hence, the P&R scheme is applied here in the sensitivity simulation EMAClnox. anthropogenic sources with a flash rate of up to 3 flashes km −2 a −1 (Fig. 9b). The Grewe parameterization used in EMACref calculates too high flashes over the free ocean due to the applied vertical C-shape profile as it has been indicated by Tost et al. (2007b). This is most pronounced in the western Pacific. Applying the common P&R parameterization in EMAClnox decreases the NO x lightning emissions over the Indian and the tropical Pacific oceans towards a better agreement with the observed values. Miyazaki et al. (2014) claimed that the dependency in the P&R scheme is too weak over the ocean. The flash 410 rates over the regions, such as Indonesia and South America however, are higher than the observations. The overestimation over South America has been also reported by Miyazaki et al. (2014). In literature the P&R scheme is criticised for linking the cloud top height and the electrification only indirectly (Tost et al., 2007b). The changes of the lightning activity correspond to a 20-30 % increase of the tropospheric NO 2 column over land and a 40-60 % decrease over oceans. This improves the bias between EMAC and OMI data over the tropical oceans and some continental emission regions like North Africa and Central 415 U.S (Fig. 16c). The changes of NO x impact the tropospheric ozone column in the Tropics leading to a subsequent 5-10 % decrease of the annual mean O 3 over the tropical Pacific (towards a lower bias) and a small increase over the tropical continents (Figs. 12c, 13c). Globally, the P&R scheme yields 423 mol(NO) flash −1 (+32 %) whereas the common estimates range from 200 to 600 mol(NO) flash −1 Gordillo-Vázquez et al., 2019;Nault et al., 2017;Marais et al., 2018).
This corresponds to a global annual LNO x emission of 5.7 Tg(N) a −1 (as in EMACref) which agrees well with the reported 420 range of estimates (Tab. 4).
As demonstrated here, the usage of the commonly applied P&R scheme improves significantly the NO 2 bias between EMAC and OMI as well as the tropospheric O 3 discrepancy to IASI over the tropical Pacific Oceans.

Advanced representation of tropospheric ozone
The previous sections demonstrate the importance of the different model developments applied in this study, which is sum-425 marised in Table 2. We combine the implementation of these developments in the simulation EMACmulti. An exception is the inclusion of the H 2 O complexes, for which the effect of water in the RO 2 /NO reaction is neglected in EMACmulti due to the sparse evidence of kinetic data. Also, the global impact of this modification on the O x chemistry is minor (Sec. 4).

Global impact on tropospheric ozone and its precursors
Among the different sensitivity simulations EMACh2o has the largest impact on ozone at the surface and in the troposphere (up 430 to 300 hPa). Although the revised dry deposition scheme influences surface ozone significantly in some regions (Sec. 5), its overall impact in EMACmulti is minor (Fig. 10). The flux is indeed mainly driven by the decreased background concentration which arise from the modified kinetics in EMACh2o. Thus, the discrepancy between EMACmulti and TOAR surface ozone in Europe and the United States as shown in Figure 11d and Figure 11h corresponds to the comparison with EMACh2o. The remaining EMAC/TOAR bias of ± 4 nmol mol −1 lays for most regions within the measurement uncertainty.

435
Considering the troposphere below 300 hPa, two other effects contribute additionally to the overall change. On the one hand, the ozone increase caused by the inclusion of HONO soil emissions (Sec. 7) counterbalances the impact of the water inclusion in the Tropics and extra-Tropics. Over the tropical oceans, the tropospheric O 3 column is significantly reduced by the usage of the P&R lightning NO x scheme (Sec. 8). This leads to a significant reduction of tropospheric ozone by about 15 % (Fig. 13d).
The resulting bias of up to 6 DU represents an improvement against the former reported EMAC bias towards satellite retrievals 440 15 DU reported by Jöckel et al. (2016). The remaining mismatch exceeds the uncertainties on the IASI retrieved tropospheric O 3 column of 5-20 %. The remaining O 3 bias over the oceans is probably associated with an underestimated ozone loss by halogens (Sherwen et al., 2016) and HO x -chemistry in clouds (Rosanka et al., 2021b).
Considering the total tropospheric O 3 column, the changes by EMACmulti (Fig. 15b) are even stronger (down to -30 %).
Here a large reduction is seen over the southern polar regions, in contrast to the comparison of the smoothed model data 445 (Fig. 13). This is due to the lower sensitivity of IASI over the polar regions (low brightness temperature) characterised by the applied averaging kernel (see Sec. 3.2). Figure 14 show that the induced changes reach up the tropopause at all latitudes, the largest in the extratropical southern hemisphere and in the tropical upper troposphere. Besides the important impact on tropospheric chemistry, which can be fully discussed here, the O 3 decrease modify the radiative forcing of tropospheric ozone.
In fact, Stevenson et al. (2013, and references therein) have estimated a change of ∼40 mW m −2 total radiative forcing per 450 one DU. The radiative effect of pertubations per unit mass of ozone is maximum in the tropical upper troposhere where our revised model predicts changes of -10 % or lower (Riese et al., 2012).
Tropospheric NO 2 overall is decreased by EMACmulti (Fig. 16d). The decreased NO 2 levels from EMACh2o are balanced and partly compensated by the increase of tropospheric NO 2 due to EMAChono, the strongest in Central Africa and Australia (Sec. 7). Globally most important are the reduced NO 2 levels over the oceans which lower the model overestimation in the 455 tropical Pacific and Indian Ocean but cause a negative bias in the Northern Atlantic Ocean. An underestimation over South Africa and East Asia exceeding the OMI bias (1 × 10 15 molecules cm −2 ) remains. Since South Africa is a region with high biomass burning emissions the bias which is highest in boreal summer and autumn (biomass burning seasons) might be due to an under-predicted fire emission factor in this area. In contrast, the underestimation over East Asia can be attributed to anthropogenic emissions which is the highest NO x source in this region. Overall, however, the noted discrepancies are within 460 the relative uncertainties of NO x emissions (Sec. 10).
Considering the calculated O x budgets, the chemical production and loss estimates by EMACmulti match better the most recent multi-model estimates of 4500-5200 Tg a −1 chemical production and 4000-4800 Tg a −1 loss where EMACref gives to high estimates. In contrast, the global dry deposition estimate by EMACmulti is much lower than the recent multi-model mean estimate of 1000 Tg a −1 (Young et al., 2018). This might point to an under-represented dry deposition (e.g. at soil) in the 465 model (Sect. 5). But it has to be noted that the calculated global dry deposition flux depends on the background concentration of ozone and therefore the decreased O x dry deposition is a consequence of the reduced net production (Tab. 3).
10 NO x emissions and the uncertainty NO x is the major precursor of tropospheric ozone driving the chemical production and loss (Monks, 2005). In low and moderate where insufficient ground measurements lead to a lack of spatial and variable information (Andela et al., 2013). A further high uncertainty is likely due to undetected small fires (>100 hectare) whose fraction Ramo et al. (2021) estimated to 80-100 % over Africa. The MODIS instruments on the Aqua and Terra satellites, which are among others the origin of the GFAS inventory, lack the observations of small fires. Especially, towards the equator where the overpasses of the polar orbiting 480 satellites reduce to two days. Moreover, the fire radiative power is underestimated by GFAS in the Tropics since the average of daily observation does not account for the high oscillation during day in these areas (Andela et al., 2013). This likely reasons the absolute difference between the annual mean tropospheric NO 2 column of EMAC and OMI over southern Africa, as it have been also reported by Andela et al. (2013). In this region, the most widerspread fires occur during boreal summer and autumn (see NASA Earth Observatory, last access: 8 June 2021) where the reported EMAC/OMI NO 2 bias is highest. A further 485 contributing reason for the NO 2 underestimation could be that peat fires which also occur in this region and emit low amounts of NO x and high amounts of VOCs are hardly estimated in bottom-up approaches, such as GFAS (Krol et al., 2013).
It has been shown that anthropogenic NO x emissions have an uncertainties of at least 20 % (Solazzo et al., 2021;Ding et al., 2017). As these emissions account for roughly 65 % of the total NO x emissions (Pozzer et al., 2012), anthropogenic sources can have a large contribution on the overall uncertainties, especially over the North Hemisphere where most of these emissions 490 are located.
Lightning activity is the major source of NO x over remote oceans accounting for 10-30 % of the global emissions Kang et al., 2020). The regional distribution of lightning is highly uncertain, as it has been shown by Tost et al. (2007b) applying different convection and lightning parameterizations in the EMAC model. Especially, in the tropics the representation of cumulus convection is challenging . Further uncertainty is attributed to the non-linear 495 NO x photo-chemistry which can only be captured in high-resolution models. Namely, NO 2 is converted rapidly to NO y and NO z (e.g. HNO 3 ) in remnant NO x plumes after thunderstorms (Nault et al., 2017). Observational constraints take this nonlinear NO x photo-chemistry (plume chemistry) into account and are used to adjust the NO production per flash and thus tune emissions parameterizations. However, this plume chemistry is a sub-grid scale process for most global atmospheric models and is commonly not parameterized. Representation of the plume chemistry for NO x by lighting has been shown to yield 500 significant changes in the O 3 levels predicted by a global model (Gressent et al., 2016). In addition, the recent study by Brune et al. (2021) has reported relevant HO x production by lightning activity which also would explain a high NO x -to NO z conversion.
2019). Here, we have implemented or applied important features in EMAC influencing the relationship between ozone and weather. Namely, the formation of water-complexes in reactions of hydroxyl and hydroperoxy radicals with nitrogen oxides is enabled affecting the tropospheric ozone chemistry. For dry deposition at vegetation we apply a parameterization extended with additional meteorological stress factors for the stomatal uptake and a weather-dependent explicit formulation of cuticular uptake. In addition, soil emissions of nitrous acid, the major precursor of the OH radical, are represented for the first time in EMAC. Also, important for tropospheric ozone, gaining more relevance in the light of global warming, is the here considered drought-dependence of biogenic volatile organic compound emissions. Finally, we investigate the production by lightning activity, a relevant source of nitrogen oxides, using a parameterization with a better distinction of land and ocean.
A detailed analysis of the separate impacts by means of the tropospheric odd oxygen budget and the subsequent comparison with ozone measurements at the surface (TOAR) and for tropospheric columns (IASI) have been conducted. In addition, the 515 impacts on nitrogen dioxide have also been investigated with respect to the OMI QA4ECV product. The inclusion of the model developments overall reduces the bias between simulated and measured ozone abundances. In fact, the comparison with TOAR data at ground level results in a remaining mismatch of ± 4 nmol mol −1 (during day). This mainly arise from the inclusion of HO 2 -water complex in the HO 2 + NO reaction and the enhanced cuticular deposition. This represents an improvement against the reported ACCMIP multi-model ensemble bias to TOAR data (Young et al., 2013).  (Sherwen et al., 2016). These expected reductions, however, will be partially compensated by NO x -recycling resulting from heterogeneous chemistry leading to ClNO 2 (Riedel et al., 2014) and HONO (Benedict et al., 2017), which are not included in the EMAC model yet. The overall reduction of tropospheric NO 2 mainly leads to solving the model overestimation over the tropical oceans. Overall, the bias is within the uncertainties of the NO x emissions.
The NO soil emission representation in EMAC, which do not present a contiguous dependence on soil moisture and tem-530 perature yet, will be updated with the parameterization of Hudman et al. (2012). Such parameterization also represents pulsing of the emissions following dry spells and N-inputs from chemical fertilizer and atmospheric N-deposition, yielding 34 % more annual global soil emissions of nitrogen oxide with a higher contribution from fertilisation. For emissions of biogenic volatile organic compounds the implementation of MEGAN3  in EMAC will lead to some improvements, by allowing at least a consistent online calculation of the drought emission activity factor, which we showed to have a non-negligible 535 impact on total ozone column in the model. Finally the use of a more up-to-date anthropogenic emissions database for the studied year would improve the results against the observational dataset, especially over polluted regions. For example, the EDGAR (Emissions Database for Global Atmospheric Research, version 4.3.2) do have 8 % larger global anthropogenic NO x emissions for the year 2010 than the one used here, implying a much larger difference at regional level. The here investigated weather-dependent processes such as dry deposition will be considered in simulations at much higher spatial resolution 540 for studying air pollution during extreme heat and drought events, which are known to increase in frequency during global warming.      Oswald et al. (2013) to The International Geosphere Biosphere Programme (IGBP) land cover scheme (Belward, 1992) No. IGBP category soil samples 1 forests S1 -S4 2 crops S10-15, S17 3 woody savanna S6/7 4 grassland S5, S8/9