LES study of the impact of moist thermals on the oxidative capacity of the atmosphere in southern West Africa

. The hydroxyl radical (OH) is a highly reactive specie and plays a key role in the oxidative capacity of the atmosphere. The total OH reactivity, corresponding to the inverse of OH lifetime, may have a signiﬁcant fraction non-attributable to commonly measured compounds. The turbulence-driven segregation of OH and its reactants can cause substantial modiﬁcation of averaged reaction rates, and thus of the total OH reactivity, when compared to a perfectly mixed assumption. We study the impact of turbulent mixing on the OH reactivity with Large-Eddy Simulations from the Meso-NH model coupled on-line with a 5 detailed chemistry mechanism in two contrasted regimes. Our ﬁndings show that the non-mixing of isoprene (resp. aldehydes) and OH leads to 30% decrease (resp. 16% increase) of the mean reaction rate at the top of the boundary layer and consequently to 9% decrease (resp. 5% increase) of the OH total reactivity in a biogenic (resp. anthropogenic) environment. Moreover, the total OH reactivity is highest inside thermals in both cases.


Introduction
The hydroxyl radical (OH) is an efficient cleansing molecule present in the troposphere. It is mainly produced during daytime through the reaction of water vapor with O( 1 D), formed by the ozone photolysis, while nitrogen oxides and volatile organic compounds (VOC) are its major sinks. OH has a very high reactivity and reacts with numerous chemical species, controlling their chemical lifetimes (Ehhalt, 1999). Several field campaigns have been conducted to study the total OH reactivity for urban 15 or forested environments. The measured reactivity have been compared to the calculated reactivity, obtained by the sum of OH reactants concentrations multiplied by their reaction constants. A missing part, corresponding to the difference in measured and calculated OH reactivity, is found under urban or biogenic conditions but also in clean remote regions. OH reactivity measurements made in urban areas proved to be in agreement (less than 10%) with the calculated OH reactivity in New York (Ren, 2003), in Houston (Mao et al., 2010) or in a urban controlled environment (Hansen et al., 2015b). However, discrepancies 20 in urban areas have been noticed between the measured and the calculated OH reactivity in Nashville (Kovacs et al., 2003) (35% less for the calculated reactivity), in Mexico (Shirley et al., 2006) (25%) or in Tokyo (Sadanaga, 2004) (25%). The differences between the measured and the calculated total OH reactivity are higher in forested areas. Di Carlo (2004) found a missing fraction of 50% in OH reactivity during the PROPHET campaign in Michigan. These results are comparable to the missing part (50 to 58%) calculated from measurements made in a boreal forest in Finland (Sinha et al., 2010;Nölscher et al., 2012). Similarly, Nölscher et al. (2016) calculated an accounted fraction of OH reactivity close to 49% in an amazonian rainforest. 5 As addressed by Williams and Brune (2015), atmospheric models do not correctly simulate the observed total OH reactivity.
Attempts to use box models with detailed chemical mechanism or 3D-models with coarse resolution to fill the gap in OH reactivity proved to be insufficient but the reasons still remain unclear. This could be partly due to not yet discovered OH reactions pathways, and thus not implemented in atmospheric models. OH recycling during the isoprene oxidation chain in forest environments characterized by lox NO x conditions was proposed to explain the uncertainties in the simulated HO x 10 budget (Lelieveld et al., 2008;Butler et al., 2008;Peeters et al., 2009;Pugh et al., 2010;Stone et al., 2011). However, Stone et al. (2011) studied several proposed OH recycling mechanism present in the literature and found that biases from OH and HO2 concentrations still exist whatever the mechanism. One possible issue in the total OH reactivity retrieval not mentioned by previous studies could lie in neglecting the turbulent motions in the transport of chemical compounds in the boundary layer, which can segregate chemical species and reduce the mean reaction rate. 15 The atmospheric boundary layer has a turbulent structure characterized by strong and narrow updrafts surrounded by weak and large descending areas (Molemaker and Vilà-Guerau de Arellano, 1998;Schumann, 1989). Considering passive scalars, Wyngaard and Brost (1984) found that the surface bottom-up transport is more important to the vertical diffusion than the entrainment zone top-down transport in a convective boundary layer. Updrafts in the boundary layer create a spatial discrimination of pollutants concentrations between thermals and their environment. This heterogeneity in chemical species redistribution in- 20 fluence the mean reaction rate obtained considering averaged reactants concentrations (Vilà-Guerau de Arellano and Cuijpers, 2000). By an idealized simulation, Molemaker and Vilà-Guerau de Arellano (1998) have shown that, for a second-order reaction implying a top-down and a bottom-up species, reaction rates are maximum in updrafts near the surface and in downdrafts at the top of the boundary-layer. Ouwersloot et al. (2011) studied the inefficiency of turbulent mixing over heterogeneous surfaces and found that the highest reaction rates for isoprene and OH are located in thermals at the top of the boundary layer. The 25 chemical reactivity of boundary layer is then determined by the ability of turbulence to mix reactive species (Molemaker and Vilà-Guerau de Arellano, 1998).
Vertical motions associated with clouds and sea-breeze impact the atmospheric chemistry and pollution levels near the surface since they dilute the chemical species and increase the upward transport of surface emissions (Chen et al., 2012). When updrafts lead to cloud formation, Vilà-Guerau de Arellano et al. (2005) found a decrease of 10 to 50% of tracer mixing ratios High resolution simulations which explicitly resolve the turbulent and convective advection terms were conducted (Vilà-Guerau de Arellano and Cuijpers, 2000;Vilà-Guerau de Arellano et al., 2005;Kim et al., 2012Kim et al., , 2016 in order to asses the impacts of clouds and convective boundary layer on the mixing of chemical compounds. However, previous numerical studies on the impact of turbulent mixing of chemical compounds used mainly rather simple or little more detailed chemical schemes, resulting in possible limitations in the representation of the atmospheric chemistry. 5 The goal of this work is to evaluate the role of thermals on OH reactivity in the framework of a convective boundary layer with contrasted chemical environments in southern West Africa. This region is characterized by high occurrence of low-level stratus and stratus deck poorly represented in climate models, leading to errors in radiative forcing (Schrage et al., 2007;Knippertz et al., 2011;Schrage and Fink, 2012). Two chemical regimes represented by a detailed chemical scheme are contrasted in LES studies. The first simulation is influenced by biogenic emissions whereas the second one is characterized by  (Lafore et al., 1998). Cloud microphysical processes are represented by the ICE3 scheme (Pinty and Jabouille, 1998) that includes six different types of hydrometeors. The turbulence is solved by a 3-dimensional scheme using a prognostic equation for the turbulent kinetic energy (Cuxart et al., 2000) with the turbulent mixing length given by the mesh size. Surface processes 20 and interactions with the atmosphere are simulated by the SURFEx model (Masson et al., 2013) coupled with Meso-NH.
The resolution used is 50m * 50m for a domain size of 10 km × 10 km (200 × 200 grid points); the boundary conditions are cyclic. 10 km is the targeted mesh size of an increasing number of current large-scale chemistry models. Along the vertical, 121 levels are stretched from ∆z = 20 m at the surface to 250 m on top of the domain at 20 km.
The simulation is run for three days whose two days consist in the spin-up for chemistry. Each day uses the same dynamical 25 conditions and has the same initial and forcing thermodynamical fields. Results are shown only for the third day from 0600 to 1700 UTC, when the convective boundary-layer is well developed. A passive scalar is emitted only during this part of the simulation with a constant emission rate and is used to determine the boundary layer height (see section 2.4.1). Thermals are identified by the conditional sampling method implemented in the model by Couvreux et al. (2010), which relies on a radioactive-decay passive tracer mixing ratio and vertical velocity anomalies. This passive tracer is emitted starting from the 30 beginning of the third day of simulation.
Initial and forcing dynamical fields come from Couvreux et al. (2014) who studied by means of a single column model the representation of the diurnal cycles of meteorological parameters in four observation sites in West Africa. Here we focus on the "cloudy" regime of Couvreux et al. (2014), representative of the climate encountered close to the Guinean Gulf. The vegetation present in our simulation is dominated by tropical crops and open shrublands (35% of the domain for each type), sea (15%), inland water (5%), wetlands (5%) and tropical grasslands (5%) and a high moisture content is prescribed with a soil water index of 0.7 and 0.74 for the surface and the ground, respectively. Initial conditions and composite large-scale advections are extracted from ECMWF re-analysis (Agustí-Panareda et al., 2010) prepared for the AMMA campaign (Redelsperger et al., 2006). The 5 present simulation is initialized at 0600 UTC with stable initial conditions, extracted from ECMWF AMMA reanalysis (black curve in Fig. 1b). The large-scale conditions from the ECMWF re-analysis also includes sea-breeze circulations from the surface to 500 m, linked to moist and cool advection throughout the simulation, topped by advection of dry and warm air from 1000 to 3000 m.
2.2 Chemical model setup 10 The applied chemical scheme ReLACS 3.0 (Reduced Lumped Atmospheric Chemical Scheme version 3.0) (Tulet et al., 2006), is a reduced version of the Caltech Atmospheric Chemistry Mechanism (CACM) (Griffin, 2002). This mechanism describes the reactions system of ozone gaseous precursors as well as of Secondary Organic Aerosol (SOA) with 365 reactions involving 87 species.
For both simulations, the initial vertical profiles of main primary chemical species are taken from airborne measurements 15 made during the B235 flight of the AMMA campaign performed by the BAE-146 over a tropical forest (Table 1).
Biogenic emissions (Table 2) are taken from the MEGAN-MACC (Model of Emissions of Gases and Aerosols from Nature -Monitoring Atmospheric Composition and Climate) inventory (Sindelarova et al., 2014) except for NO. A maximum value of 10 ng.N.m −2 .s −1 is prescribed for nitrogen oxide emissions by soils in the simulation. Biogenic NO x emissions from GEIAv1 (Global Emission InitiAtive) inventory (Yienger and Levy, 1995) were proved to be too low for the studied region in 20 comparison to estimations based on airborne measurements during the AMMA campaign (Stewart et al., 2008;Delon et al., 2010).
The emissions are constant in space and time except for NO, isoprene and monoterpenes, for which a gaussian diurnal cycle is used. For biogenic NO, the maximum emission is at 1300 UTC and the standard deviation of the gaussian curve is equal to 3 hours, these parameters are set in order to approximate the ground temperature as Mamtimin et al. (2016) and Yienger and 25 Levy (1995) noted that NO emissions from soils are closely linked to soil temperature. For isoprene and sum of monoterpenes (represented as ISOP, BIOL and BIOH in the chemical scheme), the maximum is at 1200 UTC and the standard deviation is equal to 2.5 hours, chosen to fit the diurnal evolution of the solar radiation reaching the surface. This is in agreement with mixing ratios. This affects especially long-lived species in the boundary layer and leads to the deletion of the biogenic emissions signature. Values of anthropogenic emissions are representative of Cotonou (Table 2) (Junker and Liousse, 2008).
In the following, the study focuses on isoprene for the biogenic case because it is the major biogenic VOC emitted to the atmosphere and influences the ozone and secondary organic aerosol formation (Guenther et al., 2006). For the anthropogenic case, a focus is made on the lumped C>2 aldehydes (ALD2 in the chemical scheme) because they have both primary and 5 secondary sources and their role is very important in the troposphere as they contribute to the production of radicals and are precursors of ozone (Williams et al., 1996).

Metrics
In order to study the competition between chemical reactivity and turbulent mixing, as well as the inhomogeneity in chemical species mixing, Schumann (1989) introduced two dimensionless numbers, the Damkhöler number and the segregation coeffi-10 cient. The first one corresponds to the ratio between the turbulence characteristic timescale τ turb and the chemical reactivity timescale τ chem . For a given compound A, the Damkhöler number D a is : Where w * and h refer to the convective velocity and the boundary-layer height, r A is the mixing ratio of A and sinks(A) corresponds to the total chemical loss rate of A. Schumann (1989) used a threshold value of 0.1 for the Damkhöler number discriminate slow and fast chemical reactions and this method will be used in the following. If D a < 1, the turbulent mixing is more efficient than the chemistry. If D a 1, one can expect strong competition between dynamics and chemical reactivity.
Finally, if D a > 1, the chemical reactions occur faster than turbulent mixing. This turbulent mixing causes fluctuations of 20 chemical specie mixing ratios in the LES domain, which can be quantified by the intensity of segregation. For a second order reaction implying two species A and B with a reaction constant k : A + B→C, the intensity of segregation I S (A, B) is defined as : The lower case letters represent species mixing ratios at a grid point. The overbar denotes a spatial average, and the prime a 25 deviation from this average. If I S (A, B) = −1, then the two species are completely segregated and no reaction will take place between them. If I S (A, B) = 0, compounds are perfectly mixed. A positive segregation coefficient means that covariance between species is similar and thus the mean chemical reaction rate would be higher in comparison to a perfect mixing. The segregation is calculated with mixing ratios anomalies related to spatial averages in numerical models. Ouwersloot et al. (2011) stated that these averages should correspond to the complete mixing volume in order to allow comparisons between models and measurements. In the following, the large-scale spatial average is calculated over the 10 km x 10 km model domain. One can define an effective average reaction rate R that includes the turbulent mixing impact on the chemical reactivity as : Here, k e is the effective mean reaction constant and bars over species denote averages over the domain.
In the current LES experiments, chemical reaction rates are calculated with a focus on OH. The OH radical reactivity R OH 5 corresponds to the inverse of OH lifetime τ OH and is defined as : In Equation 4, k (χi+OH) represents the reaction constant between OH and the i-th reactant, and χ i corresponds to its concentration. In a similar way than the effective reaction constant R e , an effective reactivity for the OH radical R e OH is defined by including the effect of turbulent mixing in Equation 4 as : R OH denotes the OH total reactivity calculated with averaged values. In order to obtain the relative deviation of the total OH reactivity from the reactivity computed with averaged mixing ratios, a factorization is performed on Equation 7, which results 15 in : From Equation 8 the mean relative error, E k OH , made on the total OH reactivity only considering averaged values is : The segregation intensity used to compute the mean error corresponds to deviation from boundary layer averaged values. 20 This error is not considered in previous numerical studies looking at the identification of the missing part of OH reactivity. Indeed, using a box model or a single column model like Mogensen et al. (2011);Whalley et al. (2011Whalley et al. ( , 2016 could lead to neglecting the turbulent motions that could affect the redistribution of chemical species within the atmospheric boundary layer.

Dynamics
The diurnal evolution of the boundary-layer height (BLH) is analyzed in Fig. 1a. It is diagnosed from two different methods.

5
The first one corresponds to a determination according to the bulk Richardson number method as presented in Zhang et al. (2014). The boundary layer height is defined as a threshold value for the bulk Richardson number Ri b , which is computed at a given height with the virtual potential temperature θ v and horizontal wind speeds u z and v z at this altitude and at the surface : A clear diurnal cycle is observed, the maximum height being at 1400 UTC and the minimum during nighttime (Fig. 1a, red line). the height at which a passive bottom-up scalar emitted the third day is equal to 0.5% of its surface value. Kim et al. (2012) mentioned that this definition could be valuable when studying the boundary layer deepening due to clouds. This diagnostic captures the growth of the boundary layer height due to the development of cumulus clouds (Fig. 1a, blue line) and is used to 20 mark the BL height in the following.
The range of simulated virtual potential temperature ( Fig. 1b) slightly overestimates the AMMA observations in Cotonou of Gounou et al. (2012) in the lowest 500m as shown by Couvreux et al. (2014). Both the dry and cloudy layers are identified by two inflections on the vertical profiles of virtual potential temperature (Fig. 1b). The first inflection increases from 400 m at 0600 UTC to 600 m at 1700 UTC corresponds to a thin inversion zone between the well-mixed dry layer and the cloudy layer 25 above it. The second inflection defines the top of the boundary layer ranging from 500 m at 0600 UTC to 1600 m at 1700 UTC.
Cloud cover over West Africa is an important feature of the African monsoon but is poorly represented by global models (Hannak et al., 2017). This could lead to too low simulated clouds and too high incoming radiation at the surface that imply too high diurnal cycles of temperature and relative humidity over this region. The nocturnal low-level stratus have been studied during the monsoon period at Parakou (Benin) by Schrage et al. (2007) with radiosondes. The authors found that turbulent 30 processes are responsible for cloudy nights whereas clear nights are associated with a nocturnal inversion promoting the decoupling of surface and lower atmosphere. Schrage and Fink (2012) investigated the nighttime cloud formation. They stated that the presence of a nighttime low-level jet induces a shear-driven vertical mixing of moisture accumulated near the surface.
This leads to cloud formation whose cover is likely to persist until the early afternoon (Schrage et al., 2007;Schrage and Fink, 2012). This feature is simulated in this work as low level stratus clouds occupy a large fraction of the simulated domain in the morning ( Fig. 1a and 2a;2c). In the afternoon, the cloud deck breaks up and less uniform but higher cumulus clouds are simulated (Fig. 1a, 2c and d). 5 Throughout the growth of the boundary layer, dry thermals develop capped by the temperature inversion zone. Some thermals penetrate this inversion (Fig. 2) and cloud formation occur at the upper part of the updrafts. The vertical profiles of the fraction area ( Fig. 3) occupied by thermals exhibit a local maximum at the height corresponding to the separation between the two layers. This altitude is variable through the simulation but tends to stabilize between 500 and 600m in the early afternoon and is associated to a favored detrainment zone. The peaks observed at the surface and at the top of the cloudy layer correspond to 10 two others local maxima. Due to the deepening of the boundary-layer, the top of the cloudy layer is increasing throughout the simulation ( Fig. 2,3). Simulated convective velocity w * (resp. turbulence characteristic timescale τ turb ) ranges from 1cm.s −1 (resp. 6h) in the morning to 1m.s −1 (resp. 20 min) at midday.

Atmospheric chemistry
Typical diurnal cycles are obtained for isoprene in the biogenic case and for OH in the biogenic and anthropogenic cases, where 15 both compounds exhibit maximum mixing ratios around midday (Fig. 4). For the biogenic case, the simulated isoprene mixing ratios averaged from the surface to 600 m reach a maximum close to 1 ppbv ( Fig. 4a) at noon, when isoprene emissions are the highest. It is on the same range as AMMA measurement studied by Saxton et al. (2007) who found a maximum of 1.5 ppbv on the composite diurnal cycle of isoprene at midday. In our study, the ozone mixing ratios do not exhibit strong variability throughout the day. The simulated values around 18 ppbv are lower than observations collected over forested areas (Table 3) 20 ranging from 22 to 30 ppbv. NO x mixing ratios are close to 0.2 ppbv on average around midday (Fig. 4b) and are in agreement with AMMA measurements where a mean value of 0.1 ppbv of NO x has been observed. Simulated OH mixing ratios vary between 0 and 0.18 pptv (Fig. 4a), within the observational range of 0.05 to 0.15 pptv during the AMMA campaign (Table 3).
The biogenic environment is characteristic of a NO x limited regime.
The chemical regime induced by the anthropogenic emissions is contrasted with the previous biogenic case (Fig. 4c,d), 25 including NO x emissions nearly forty times higher at 1300 UTC. Averaged NO x mixing ratio is 1.5 ppbv at midday (Fig. 4d), which is similar to the averaged value of 1 ppbv measured during a flight over Cotonou during the AMMA campaign (Table   3). The increase in NO and NO 2 leads to important oxidant formations. OH mixing ratios vary between 0 and 0.40 pptv in the simulation, in agreement with AMMA measurements ranging from 0 to 0.50 pptv over Lagos (Table 3). Ozone is produced throughout the simulation but does not exhibit strong spatial variabilities. Its mixing ratios vary from 17 ppbv in the morning 30 to 82 ppbv in late afternoon. This is higher than observations collected over several cities in West Africa varying between 26 and 40 ppbv (Table 3). ALD2 is continuously increasing during this case through chemical production and emissions (Table   2). Its mixing ratios range from less than 1 ppbv in the morning to 11 ppbv at the end of the simulation. Unfortunately, no C>2 aldehydes observations, except formaldehyde, were available from the AMMA experiment despite the use of a PTR-MS due to interferences during the measurements for m/z=45 identified as acetaldehyde .
A minimum of NO is found around 1200 UTC for both cases. At that time, NO is efficiently converted into NO 2 . However, NO 2 chemical transformations in reservoir species such as peroxy pentionyl nitrate (PAN1 in the chemical scheme), peroxy acetyl nitrate (PAN2) or HNO4 are net sinks for NO 2 . The chemical balance between reservoir species and NO 2 represent 5 2.12% of the net destruction of NO 2 averaged over the domain at 20 m and 1200 UTC for the biogenic case, and 34.2% for the anthropogenic case. As a consequence, less NO 2 is available to convert into NO, which explain the midday low NO mixing ratios ( Fig. 4b and d).
3 Vertical transport and chemical reactions in the convective boundary layer 3.1 Impact of turbulent mixing on the OH reactions 10

Vertical transport and Damkhöler numbers
Isoprene is highly reactive, especially towards OH, and is rapidly consumed in the boundary layer linked to its chemical lifetime τ ch approximately equal to 30 mn in the biogenic case. The Damkhöler number of isoprene is close to 1 (Table 4), indicating that the isoprene chemical lifetime is comparable with the timescale of turbulent mixing. This implies that chemical reactions and turbulent mixing are competing processes for this compound. Isoprene is both transported and consumed inside 15 the thermals and exhibit vertical and horizontal gradients (Fig. 5a). Average profiles of isoprene are decreasing with altitude ( Fig. 5a). The lower mixing ratios on the dashed line, denoting updrafts-free region, are close to the domain averaged values due to the domain coverage by thermals (Fig. 3). Updrafts regions contain isoprene higher mixing ratios and anomalies are consequent over the whole boundary layer.
For the biogenic case, OH has a very short chemical lifetime of the order of 0.2 s. The OH radical reaches rapidly a 20 chemical equilibrium and is nearly undisturbed by the turbulent mixing as shown by its large Damkhöler number (Table 4). As a consequence, OH mixing ratios (Fig. 5b)  isoprene mixing ratios are the highest. This implies that both the surface and air lifted by thermals are preferential reaction zones in the boundary layer.
For the anthropogenic simulation, ALD2 reaction rate with OH is lower than isoprene. Its calculated Damkhöler number (≈ 0.17) indicates that turbulent mixing dominates over chemical reactions (Table 4). Consequently, ALD2 is efficiently transported by updrafts (Fig. 5c). The contrast between concentrated air parcels lifted up by thermals and relatively diluted air 30 outside is illustrated by the large differences in the ALD2 average profile in thermals and in the environment (Fig. 5c). Sim- ilarly to isoprene in the biogenic case, two local maxima are present 600 and 1100m, representing the two-layer structure presented above.
OH mixing ratios are nearly twice higher in the anthropogenic environment than the biogenic case. That induces a more reactive atmosphere and a lower chemical lifetime for species whose OH is the main sink in the boundary layer. OH has a very low chemical lifetime of 0.07s in this simulation, which represents a large Damkhöler number close to 12 000. OH mixing 5 ratios are maximum in thermals (Fig. 5d) due to the transport of OH precursors, such as NO x , and fast chemical equilibrium.
Conversely, lower mixing ratios on OH average profiles (Fig. 5d) corresponds to regions without updrafts, leading to strong OH anomalies inside updrafts.

Vertical profile of segregation intensity
Negative values of segregation coefficient up to -30% are calculated at the top of the cloudy boundary layer from 1000 to 1700 10 UTC which means that OH and isoprene are partly segregated in this frontier zone. In other words, a well mixed atmosphere hypothesis would conduct to a 30% overestimation of the reaction rate at the frontier between the boundary layer and the free troposphere. The negative segregation means covariances of isoprene and OH are of opposite signs (Eq. 2). This is due to lower

15
In the biogenic case, isoprene anomalies in thermals are important from the surface (+0.48 ppbv on average at midday) to the top of the boundary layer (+0.1 ppbv on average at midday) and are thought to be always positive as it is uniformly emitted at the ground (Fig. 5a). On the opposite, OH mixing ratios are almost constant in the boundary layer at 1200 UTC (Fig. 5b), so the magnitude of its anomalies are expected to be low. Besides, its very short chemical lifetime induces that OH is quickly in equilibrium with its surroundings, which implies that its fluctuations are mostly due to thermals transporting air originating 20 from a different chemical environment. Thus, isoprene anomalies are thought to be the major driver in segregation magnitude over the boundary layer whereas OH anomalies are related to the segregation sign.
Positive values around +5% are calculated at 700 meters starting from 1400 to 1800 UTC (Fig. 6a). The segregation intensity become positive due to positive anomalies of both compounds. Due to the decreasing isoprene emissions in the afternoon, OH destruction slows down, especially inside thermals. They are still active to transport enough NO to react with HO 2 to produce 25 OH, which induces higher OH mixing ratios inside updrafts.
Before 0900 UTC near the surface, the segregation coefficient in the anthropogenic simulation between OH and ALD2 is negative up to -8% in the lower 200m (Fig. 6b) oxidation reaction by OH is accelerated up to 16% at the top of the cloudy layer from the morning to the early afternoon compared to a perfect mixing assumption.
Regarding the two simulations, the segregation has both spatial and temporal variations. The maximum values of segregation coefficient are calculated near the top of boundary layer. Below and during daytime, considered species are well mixed for both cases. It means that the highest decrease (resp. increase) induced by thermals of isoprene (resp. ALD2) + OH reaction is located 5 near the boundary layer top.

OH budget and reactivity in a convective boundary layer
The precedent part emphasizes the non uniform mixing between isoprene and OH for the biogenic case and between OH and ALD2 for the urban case, and the implication for the reaction between these species. However, this feature has to be accounted for every OH reactants in order to have the full picture of the total OH reactivity and understand how the Meso-NH model 10 compute the OH budget in different chemical regimes.

OH budget in thermals versus environment
In order to identify and quantify the major OH sources and sinks in the boundary layer, the instantaneous chemical budget of OH at 20 meters above ground level is investigated at 1200 UTC for both environments (Fig. 7). This height is the first model level and allows a comparison between the results from the simulations and from measurements in the literature. The budget 15 distinguishes between updrafts and updrafts-free columns. Percentages are related to the fraction of the overall production and destruction in or outside thermals. In the biogenic case (Fig. 7a), the OH budget is dominated by its destruction by isoprene oxidation (41.7 % of the total loss in updrafts and 29.3 % of the total loss in the rest of the domain) and its production by peroxy radicals reaction with HO 2 (37.2 % of its total source in updrafts and 42.4 % of the total source in non updrafts) and NO reaction with HO 2 (32.4 % of the total source in updrafts and 17.8 % of its total source in the rest of the domain). In that 20 case, the peroxy radicals are mainly formed by the oxidation of isoprene and its degradation products. The absolute value of OH reactivity is higher in thermals than in the rest of the domain.
The OH budget for the anthropogenic case (Fig. 7b) shows that the chemical reactivity is higher inside thermals at the surface compared to the rest of the domain. The budget is largely dominated by the production of OH by NO+HO 2 and by O 1 D+H 2 O.
Over the whole domain, ALD2+OH is the most important sink at the surface or at 500 m, followed closely by the oxidation of 25 carbon monoxide.
Similar results are obtained at 500 m (not shown) that confirm the higher chemical reactivity inside thermals compared to their environment.

OH reactivity in the convective boundary layer
The OH reactivity for the biogenic case (Eq. 4) at 20 m is maximum around midday and is equal to 6.0 s −1 , 4.25 s −1 and 4.55 30 s −1 respectively in updrafts, non-updrafts and averaged over the domain (Fig. 8a). This feature is linked to the photochemistry maximum activity at noon, and the diurnal cycle of emissions amplifies this phenomenon. At that time and near the surface, many chemical compounds are available to react with OH, which leads to a high value of reactivity. The values in updrafts are higher than outside due to higher reactants mixing ratios inside thermals (Fig. 5a). The arithmetic difference between updrafts and non-updrafts is maximum at 1200 UTC and is about 1.75 s −1 . As the thermals occupy less than 15% of the domain (Fig.   3), the evolution of the domain-averaged OH reactivity is very similar to these related to updrafts-free region.

5
The diurnal cycle of OH total reactivity with a maximum around midday in a biogenic environment is well documented in literature from observations of OH reactivity for a Mediterranean forest (Zannoni et al., 2016), for temperate forests (Sinha et al., 2008;Ramasamy et al., 2016) but also for tropical forests Williams et al., 2016). The values of OH reactivity in or outside thermals in the present study are the lower bound of measurements over forests and gathered in Yang et al. (2016), between 1 and 76 s −1 .

10
The mean relative error made on OH reactivity is calculated following Equation 9 (not shown). It is generally negative throughout the simulation, increases during the morning and is maximum in the early afternoon with a peak at -9% at 1430 UTC. In other words, neglecting the segregation of reactive species turbulent mixing in the boundary layer would overestimate the OH reactivity by 9% at the maximum in an environment dominated by biogenic emissions.
For the anthropogenic case, the total OH reactivity (Eq. 4) in thermals or in the rest of the domain does not display a clear 15 diurnal cycle (Fig. 8b). From the 0800 UTC value of 11.6 s −1 , the domain-averaged OH reactivity fluctuates but tends to increase to 14.2 s −1 at 1600 UTC. The evolution of the OH reactivity in thermals is similar and ranges from 14.8 s −1 at 0800 UTC to 17.7 s −1 at 1600 UTC. For the same period, the values in updrafts-free domain vary between 11 and 13.5 s −1 .
As a consequence of the increased OH reactants mixing ratios in the boundary layer ( Fig. 4c and d), OH reactivity is higher in this case compared to the biogenic simulation. The mean relative error of the OH reactivity (Eq. 9) is generally positive 20 throughout the simulation (not shown). It ranges between 0 to 4% from 0900 to 1730 UTC with a maximum value of 6% at 1800 UTC. In that case, the turbulent mixing induces a moderate increase up to 6% of the total OH reactivity for two reasons.
The segregation effect is limited to the last 200 m of the boundary layer. Thus averaging on the whole boundary layer suppress the extreme values. Moreover, some chemicals could have negative segregation towards OH that may compensate the positive values simulated for ALD2 and OH (Fig. 6b). 25

Discussion
The redistribution of chemical species in the boundary layer induces a different mean reaction rate between compounds when compared to a situation where chemical species would be perfectly mixed. This assumption used in regional or large scale atmospheric models leads to errors on chemical reactivity as the turbulent mixing occurs at scales smaller than the grid length (Vinuesa and Vilà-Guerau de Arellano, 2005).

30
In a biogenic environment characterized by low NO x conditions, Kim et al. (2016) found negative segregation between isoprene and OH varying between -3% near the surface to -10% in the cloud layer due to increasing OH mixing ratios in thermals with altitude. That implied positive isoprene anomalies and negative for OH in the frontier region between the boundary layer and the free troposphere, these features are reproduced in our biogenic simulation even though our values of segregation are higher in the cloud layer (Fig. 6).
Using a simple chemistry, Ouwersloot et al. (2011) found an almost constant value of -7% for the segregation between OH and isoprene over the amazonian forest in the boundary layer. It was the results of positive isoprene anomalies due to transport by thermals and negative OH anomalies due to consumption therein. Negative segregation ranging from -2% to -5% inside 5 the convective boundary layer have been simulated by Li et al. (2016) and Kim et al. (2016). In the biogenic case of our work, negative segregation of few percent in the middle of the boundary layer is reproduced. The discrepancies between this last simulation and the previous cited studies are partly due to differences between the convective boundary layer dynamics, resulting in different characteristics of turbulent structure. Moreover, using simple chemistry rather than a comprehensive chemical mechanism could lead to an insufficient representation of the OH chemistry. This results in neglecting some OH 10 destruction reactions and eventually some OH recycling pathways. Besides, Ouwersloot et al. (2011) and Kim et al. (2016) investigated the sensitivity of segregation to NO x . It was found that different NO x levels imply discrepancies in the segregation of OH and other compounds as they contribute to the OH production.
The segregations between isoprene and OH inferred from observations are higher than numerical studies calculated. Butler et al. (2008) found a segregation close to -13% over a forest in Suriname and Dlugi et al. (2010) calculated over a deciduous 15 forest a reduction of 15% in the effective reaction rate between OH and isoprene at the surface due to incomplete mixing. This is higher than the segregation computed at the surface in the biogenic case of the present study but when considering the complete mixing volume, the segregation computed with boundary layer averaged mixing ratios is negative and its maximum value is -16% at 1300 UTC. Kim et al. (2012)  formaldehyde was comparable to isoprene at that height, with HCHO mixing ratios ranging from 2 to 4.5 ppbv at the surface. Yang et al. (2016) found that in forest areas the OH budget was largely dominated by isoprene and its oxidation products.
As an example, over a Mediterranean forest, Zannoni et al. (2016) measured the OH reactivity, as well as the concentration 30 of biogenic compounds. They found that isoprene was the dominant sink for OH and contributes up to 74% of OH total reactivity during daytime due to its high reactivity towards OH and its high concentration over the forested area. The isoprene predominance in OH loss is reproduced in the OH budget of the biogenic environment of the present study, as well as for carbon monoxide. Formaldehyde mixing ratios are close to 2 ppbv, which explains that this is not a important sink as isoprene, but remains not negligible in the OH loss. The OH production in the biogenic case is similar to Kim et al. (2012) as almost the same production terms are present. However, the overall production is dominated by peroxy radicals reaction with HO 2 and the ozone mixing ratios of 18 ppbv decrease the importance of O 1 D + H 2 O source for OH.
However, the mechanism used to represent atmospheric chemistry in our simulation has a fast OH cycling through the reaction of peroxy radicals RO 2 with HO 2 . An uncertainty comes from this reaction yield that is higher than laboratory studies (Hasson et al., 2004;Jenkin et al., 2007;Groß et al., 2014;Winiberg et al., 2016). This imply an overestimation of OH mix-5 ing ratios, especially in the biogenic case where RO 2 are important. As the hydroxyl radical is recycled through the isoprene oxidation products, it tends to reduce the impact of isoprene chemistry on OH mixing ratios and thus influence the low segregation simulated in the core of the boundary layer. This is similar to a case considered by Stone et al. (2011) where a recycling mechanism ISOPO2 + HO 2 → ISOPOOH + 3 OH have been proposed to increase the simulated OH concentrations. This reaction imply that isoprene have no net impact on OH concentrations. This conclusion is drawn also by Kubistin et al. (2010) 10 who found better agreement between simulated and measured HO x concentrations by ignoring the isoprene chemistry. As the chemistry of isoprene is not well understood, the results obtained in our work are subject to these uncertainties.
Aldehydes have not been considered in previous works studying segregation. Auger and Legras (2007) calculated the segregation at 250 m and 1100 UTC between each species of the chemical model CHIMERE used in their simulation. They found a segregation ranging from 0 to -1% between OH and acetaldehyde. This is comparable to range of results of the anthropogenic 15 case of the present study (Fig. 6b) considering the incomplete mixing between OH and C>2 aldehydes at the same height. The segregation computed relatively to boundary layer averaged values is very similar to the mean relative error (Eq. 9) made on the OH reactivity due to the strong predominance of ALD2 in the OH budget. It is generally negative throughout the simulation and varies from -0.02 at 0700 UTC to -0.015 at 1730 UTC. As pointed out by the vertical profiles of segregation (Fig. 6b), the impact of turbulent mixing on the chemical reactivity depends on the considered time of the day. In this case, a reduction of 20 few percents of the reaction rate of ALD2+OH is found in the early morning and the late afternoon, and an increase up to 4% is simulated between 0800 and 1600 UTC over the whole boundary layer. This reveals a small impact of turbulent mixing on the chemical reactivity.
Several instrumental studies looked at the OH total reactivity in urban environments, and especially at the OH budget. Hansen et al. (2015a) found in an urban environment that NO x contribution (50 to 55%) was the most important to the OH 25 loss ranging considering NO x mixing ratios between 10 and 300 ppbv. The next most important contribution came from the OVOCs (especially aldehydes and ketones) that varied from 15 to 25% of the OH destruction with mixing ratios close to 20 ppbv. Some discrepancies exist between this experimental study and the OH budget for the anthropogenic case in our study (Fig. 7b). Given the NO x mixing ratios simulated in the present study (< 3 ppbv at midday), they are not an important sink for OH as measured by Hansen et al. (2015a). However, the range of OVOCs contribution reported by Hansen et al. (2015a) is 30 comparable to what is found for ALD2 in the anthropogenic case.
In the city of Tokyo, Sadanaga et al. (2004) found that the OH budget was dominated by NMHCs and by OVOCs. In their study, the OVOC category grouping acetaldehyde, formaldehyde, methanol, ethanol and acetone, can contribute up to 18% of the OH reactivity. If aggregated, the species constituting the NMHCs category are the predominant sink for OH in our anthropogenic simulation, as observed by Sadanaga et al. (2004). However, C>2 aldehydes have higher contribution in the OH reactivity missing fraction, between 5 and 15 %, are of the order of the overestimation of the total OH reactivity neglecting the turbulent mixing over a rainforest in a convective boundary layer (Fig. 8a).
Aqueous-phase chemistry was not considered here, nor the exchanges between gas and aqueous phases. However, it could have an impact on soluble species mixing ratios, such as formaldehyde and H 2 O 2 through the capture and degassing cycles of 15 these compounds. Lelieveld and Crutzen (1990) showed a decrease in oxidative capacity of the atmosphere through aqueousphase reactions via a significant decrease in ozone mixing ratios, but also HO, formaldehyde and nitrogen oxides. However, variables effects of aqueous-phase chemistry on gas-phase compounds concentrations exist (Barth et al., 2003) and OH concentrations could decrease in clouds (Mauldin et al., 1997), reducing the cleansing capacity of the atmosphere and potentially increasing the segregation between OH and its reactants. 20

Conclusions
A numerical simulation coupled with a realistic chemical mechanism is performed with the atmospheric model Meso-NH to study the impact of thermals on the oxidizing capacity of the atmosphere. The fine grid resolution of the LES version of the model allows the thermals to be explicitly resolved spatially and temporally. Identification of thermals was based on a conditional sampling method relying on a radioactive-decay passive scalar. The study of impact of turbulent mixing on the 25 chemical species redistribution, but also the consequences on the OH reactivity, is performed in a natural environment and a more contrasted urban case.
The differentiated transport by thermals is dependent of the chemical lifetime of compounds, which is represented by the Damkhöler number. This transport induces heterogeneities in the species repartition within the boundary layer, but also has an impact on the mean chemical rate between reactive species. Between isoprene and the OH radical, a 30% affective decrease of 30 the reaction rate is calculated at the top of the boundary layer in a biogenic environment compared to a perfectly mixed case.
In the urban case, the reduction of the mean chemical rate between the OH radical and C>2 aldehydes can reach 8% at the surface in the early morning while this reaction is increased up to 16% at the top of the boundary layer during most part of the simulation.
Thermals transporting species emitted at the surface can lead to different chemical regimes inside updrafts and the environment. For both cases, the surface and the transported parcel correspond to preferential reaction zones, where the reactivity is the highest. This is especially the case for the OH radical whose precursors are either transported by thermals or created inside.

5
For the natural case, the major OH precursors close to the surface are radicals originating from the oxidation of isoprene and its degradation products whereas O 1 D+H 2 O reaction become more predominant with increasing altitudes. In the urban case, OH is mainly produced through the reaction between HO 2 and NO, at the surface or higher in the boundary layer. This leads to a higher oxidation capacity in air transported by thermals for both cases. The heterogeneity in OH reactivity within the boundary layer have repercussions on the mean OH reactivity. In order to evaluate this impact, the mean relative error is introduced and 10 includes the effect of turbulence on the mean effective reaction rate between OH and its reactants. A maximum overestimation close to 9% of the total OH reactivity in the biogenic case is found compared to a reactivity that would be calculated with boundary layer averaged values. This is in the range of OH reactivity missing fraction, from 5 to 15%, found by Nölscher Couvreux, F., Guichard, F., Gounou, A., Bouniol, D., Peyrillé, P., and Köhler, M