Evolution under dark conditions of particles from old and modern diesel vehicles, in a new environmental chamber characterized with fresh exhaust emissions

Atmospheric particles have several impacts on health and environment, especially in urban areas. Part of those particles is not fresh, and has undergone atmospheric chemical and physical processes. Due to a lack of 15 representativeness in experimental conditions, and experimental artifacts such as particle wall losses in chambers, there are uncertainties on the effects of physical processes (condensation, nucleation and coagulation), and their role in particle evolution from modern vehicles. This study develops a new method to correct wall losses, accounting for size-dependence and experiment-to-experiment variations. It is applied to the evolution of fresh diesel exhaust particles to characterize the physical processes which they undergo. The correction method 20 is based on the black carbon decay and a size-dependent coefficient to correct particle distributions. Six diesel passenger cars, Euro 3 to Euro 6 were driven on a chassis dynamometer with Artemis Urban cold start and Artemis Motorway cycles. Exhaust was injected in an 8 m chamber with Teflon walls. The physical evolution of particles was characterized during 6 to 10 hours. Condensation occurs even without photochemical reactions, due to the presence of intermediate volatility organic compounds and semi-volatile organic compounds. These 25 compounds were quantified at emission, and induce a particle mass increase up to 17 %.h, mainly for the older vehicles (Euro 3 and Euro 4). Condensation is 4 times faster when the available particle surface if multiplied by 3. If initial particle number concentration is below [8-9]×10 #.cm, it can increase up to 25 %.h due to nucleation or condensation on particles below 14 nm. Above this threshold, particle number concentration decreases due to coagulation, up to -27 %.h. 30


Introduction
Air pollution is a major concern due to its impacts on climate, environment and health. It has been classified as carcinogenic to humans by the International Agency for Research on Cancer (IARC, 2016). Its effects are greater in urban areas, where pollutants such as fine particles accumulate and present a serious health risk due to human exposure. Road traffic is an important source of particles in urban environments (Rivas et al., 2020). Many 35 particle-emitting vehicles such as old diesel cars are still present on roads of several countries (EEA, 2020). https://doi.org/10.5194/amt-2021-43 Preprint. Discussion started: 8 April 2021 c Author(s) 2021. CC BY 4.0 License. due to Brownian diffusion, gravitational settling, turbulence and electrostatic charge of the Teflon walls (Leskinen et al., 2015;Nah et al., 2017;Pierce et al., 2008;Seinfeld and Pandis, 2016;Wang et al., 2018aWang et al., , 2014. Leakage and wall deposition of particles induce a decrease in particle mass over time. This prevents the analysis of the mass evolutions from condensation or evaporation of organic material onto or from pre-existing 80 particles. Moreover, since wall deposition depends on particle diameter, it affects the distribution evolutions and therefore the ability to quantify the role of coagulation. It also affects the observation of nucleation mode particles. Particle data in chambers therefore need to be corrected for leakage and wall losses. Some studies (Grieshop et al., 2009;Nah et al., 2017;Pathak et al., 2007) have used size-independent wall losses correction methods. It is usually done with the help of a tracer, such as black carbon, assumed to be reliable, and 85 enabling the consideration of the experiment-to-experiment variations. However, particle wall losses are known to be size-dependent (Charan et al., 2018;La et al., 2016;Leskinen et al., 2015;Pierce et al., 2008;Wang et al., 2018a;Weitkamp et al., 2007). The use of the same correction coefficient to the whole distribution can mislead the observation of processes such as coagulation. Other studies have therefore used size-dependent correction methods (Nah et al., 2017;Pierce et al., 2008;Wang et al., 2018a). They enable a more precise analysis of the 90 processes affecting the size distribution. However, those methods are usually based on the decay of ammonium sulphate particles (Wang et al., 2018a). They assume that their loss rates are the same as those of the studied particles. Furthermore, the loss rate studies cannot be conducted simultaneously with the actual experiments.
They are therefore usually carried out before and after the campaign, thus preventing the experiment-toexperiment variations from being accounted for (Wang et al., 2018a). 95 Firstly, this study presents the characterization of a new 8 m 3 cubic chamber with Teflon walls, aimed at studying the physical particle evolutions from a wide range of on-road vehicles. Six diesel vehicles (2 without diesel particle filter (DPF) Euro 3 and Euro 4; 3 Euro 5 with additive DPF or catalyzed DPF; 1 Euro 6 with additive DPF and selective catalytic reducer (SCR)) were tested to develop the new leakage and size-dependent wall loss correction method. Five gasoline vehicles (3 with port fuel injection Euro 3, 4 and 5; 2 Euro 5 with 100 direct injection) were also used. The new leakage and wall loss correction method is based on the complementarity of a tracer (black carbon) and a size-dependent equation to correct wall losses. It accounts for experiment-to-experiment variations.
Secondly, this study presents the emissions and evolutions of particles from the 6 diesel vehicles, Euro 3 to Euro 6, in urban and motorway conditions. Emission factors (EFs) of particle number (PN), particle mass (PM), 105 black carbon (BC), non-methane hydrocarbons (NMHCs) and IVOCs are estimated. They are used to discuss the impact of aftertreatment technologies on emissions and particle evolutions. Moreover, the results of the diesel particle physical evolutions in the chamber under dark conditions are presented. Data are corrected for leakage and wall losses using the new correction method. The physical processes are characterized in the dark in order to better understand their share in particle evolution mechanisms. This study is complementary to those focusing on 110 photochemical processes, to better evaluate the total contribution of road traffic to atmospheric particle pollution.
Dynamometer tests are performed with diesel and gasoline passenger cars (PCs), whose characteristics are given in Table 1. Six diesel vehicles are used in this study: 2 vehicles without DPF, Euro 3 (D1) and Euro 4 (D2); 3 Euro 5 vehicles, 2 with an additive DPFs (D3 and D4) and 1 with a catalyzed DPF (D5); and 1 Euro 6 vehicle (D6) equipped with an additive DPF and a SCR. Five gasoline vehicles are tested: 3 with port fuel injection (PFI), G1 to G3 with standards Euro 3, 4 and 5; 2 Euro 5 with direct injection (DI), G4 and G5. 130

Exhaust gas sampling and injection into the chamber
Exhaust gas is sampled from the tailpipe by an ejector dilutor (Dekati L7, Sapelem 50 NL.min -1 , Sapelem 135 1 NL.min -1 ). The dilutor works with air which is dehumidified, filtered using HEPA particle filters and activated carbon, and heated at 120 ° using a Fine Particle Sampler (FPS Dekati, 4000). Exhaust gas is diluted between 3 and 20 times, depending on air pressure and ejector dilutor type. The air and exhaust gas are injected into the chamber at the height of 90 cm, through a 2 to 4 m long stainless-steel line. The line is heated at 120 °C to avoid condensation and deposition on its inner surface. The heated line is cleaned after each injection with hot dry 140 clean air, during a couple of hours.

Environmental chamber
The study of the exhaust particle evolution is performed inside the University Gustave Eiffel UGE 8 m 3 cubic chamber. The structure of the chamber is made of aluminum, and the walls are made with a 70 µm thin Teflon film, used because chemically inert. However, it can be electrostatically charged and influence the wall 145 deposition of gas and particles inside the chamber. The theoretical total volume is 8000 L, but it can vary (± 300 L) due to deformation of the walls, depending on inner pressure. To avoid contamination from the outside, the chamber is always kept at a slight overpressure (~ 5 Pa) by injection of clean air. This also compensates the losses by leakage and instrument sampling. This injection induces dilution inside the chamber.
The temperature, humidity and relative pressure are continuously monitored. Instruments sample the air of the 150 chamber at the center, at 5 different heights: 20 cm, 60 cm, 100 cm, 140 and 180 cm.

Cleaning
Before the first experiment, the inner walls of the chamber were cleaned by hand with microfiber cloth. While this removes most particles and gaseous species on the walls, it also induces high electrostatic charge due to friction of the cloth on Teflon. Manual cleaning was therefore only done once. At the end of each experiment, 155 cleaning is performed with injection of dry clean air, with flowrates varying between 150 and 300 L.min -1 .
Cleaning lasts a minimum of 12 hours, thus renewing the air inside the chamber at least 13 times. Since dilution induces an exponential decrease of the pollutant concentrations, this gives a minimum theoretical percentage of clean air inside the chamber of 99 %.

Gas-phase analysis
Non-Methane Hydrocarbons (NMHCs) are measured either directly from the tailpipe with a Horiba Portable Emissions Measurement System (PEMS) or after a CVS by flame ionization detection with a Horiba analysis system.
IVOCs are collected from the heated line (prior injection into the chamber) by sampling diluted exhaust through 165 stainless steel tubes filled with Tenax TA at a flow rate of 45 mL.min -1 . The samples are collected during the entire driving cycle. Collected samples are further analyzed by Automatic Thermal Desorption coupled with Gas Chromatography-Mass Spectrometry detector (ATD-GC-MS). The Markes Unity thermodesorber and a GC6890 gas chromatograph from Agilent fitted with the MS5973 mass spectrometer from Agilent are used. The thermal desorption system consists in a 2-stage desorption. During the first desorption step, the compounds are desorbed 170 https://doi.org/10.5194/amt-2021-43 Preprint. Discussion started: 8 April 2021 c Author(s) 2021. CC BY 4.0 License. by heating the Tenax TA under a stream of helium and are condensed on a trap filled with adsorbent and maintained at -5 °C. During the second desorption, the second trap is flash-heated to 300 °C for a rapid introduction of the compounds into the chromatographic column. The chromatographic column used is an Agilent HP1MS 30 m × 0.25 mm, 0.25 µm. The mass spectrometer operates in scanning mode at an electron ionization of 70 eV. Mass spectral data are acquired over a mass range of 33-450 amu. Qualitative identification 175 of compounds is based on the match of the retention time and confirmed by matching their mass spectra with those of standards and from the NIST mass spectral library. Quantification is conducted by the external standard method. Known amount (1 µL) of standard solutions of volatile organic compounds (VOCs) and IVOCs are introduced into cleaned Tenax TA tubes, using an automatic heated GC injector. The calibration tubes are analyzed in the same conditions as mentioned before. The chromatogram shows an unresolved complex mixture, 180 mainly composed of co-eluted hydrocarbons which cannot be further separated by a single-dimensional GC. The alkanes (linear and branched) are quantified by SIR-based response factor of these compounds using the fragment of m/z=57. The response factor used for quantification of branched alkanes is the one of the compound with the same number of carbon as its parent chain. The fragment m/z=83 is used for the quantification of cyclohexane and of the other cyclic compounds. The fragment m/z=78 is used for the quantification of benzene, 185 and m/z=91 for other aromatics. The fragment m/z=128 is used for the quantification of naphthalene. The compounds from the standard solutions taken for quantification of each identified compound are listed in Table   2. The use of response factors from different compounds than the identified ones can induce some error in their quantification. Over the whole range of IVOCs, this error is estimated to be below 8 % (7.8 % and 0.4 % respectively for 2 urban cycles). For single SVOCs, the errors range from 90 to 98 %. Since all compounds 190 above C22 are quantified using compounds with different carbon numbers (C20 for vehicles D1, D5, and C16 for vehicles D2, D3, D6), the impact on their quantification is important. They are however of the same order for all vehicles, and allow comparison between vehicles and driving conditions. Linear (n-) and branched (b-) alkanes are classified according to their number of carbon C as an indicator of their effective saturation concentration, according to Zhao et al. (2015). Alkanes with C<12 are considered as VOCs, alkanes from C12 to C22 are considered as IVOCs, and alkanes with C>22 are considered as SVOCs. Also according to Zhao et al. (2015), IVOCs are considered to be the sum of n-and b-alkanes (from C12 to C22), 200 naphthalene, as well as aromatics and cycloalkanes in the same retention time bin as C12-C22 alkanes.
The CO2 concentration is measured from the chamber using a MIR2M (Environment SA), which samples air at 1.5 L.min -1 . The temporal resolution is 1 second, and the analytical uncertainty is 0.001 %.

Particle-phase analysis
Particle concentrations and size distributions are measured with a Scanning Mobility Particle Sizer (SMPS, TSI), 205 composed of an advanced aerosol neutralizer (3087), a Differential Mobility Analyzer (DMA, 3081) column, and a Condensation Particle Counter (CPC, 3775). The sampling flow is 0.3 L.min -1 . Classification is based on the particle electrical mobility, and the measurement range goes from 14 to 615 nm. The particle mass is computed, with the assumption that particles are spherical, with an arbitrary density of 1.2 (Barone et al., 2011;Totton et al., 2010). Data are given with a 5 minute time scale, with a 5 % analytic uncertainty. 210 The black carbon concentration is measured using an AE33-7 aethalometer from Magee Scientific. Air is sampled at 2 L.min -1 on filter tape. The black carbon concentration is given by absorption measurements at 880 nm (Andreae and Gelencser, 2006). Data are given with a time-scale of 1 minute.

Ammonium sulphate experiments
In order to test the wall losses correction method with non-volatile inert seeds, experiments are performed with 215 ammonium sulphate particles. Particles are generated from diluted solutions using an atomizer aerosol generator (TSI, 3079A) at a flow rate of 5 L.min -1 during 20 min to 2.5 hours. After injections, particles reached concentrations ranging from 5500 to 50600 #.cm -3 . Characteristics of these experiments are given in Table 3. chamber between 65 and 130. It is considered ideal for the sake of this study since it enables to have diluted exhaust giving particles concentrations covering a wide range, depending on emissions. This variety in initial particle concentration is important to study the influence of initial concentration on particle evolution. During injection inside the chamber, turbulence is induced due to the flow of exhaust gas, and the mixture is not homogeneous right after the end of injection. Moreover, to avoid the increase of particle deposition surface and turbulence (Crump et al., 1982;Nomura et al., 1997), there is no fan system to make the mixture homogeneous. 235 To determine the necessary time to have a homogeneous mixture inside the chamber, CO2 is injected, and the concentration is measured at 3 of the 5 sampling heights of the chamber: the bottom one (20 cm), the middle one (100 cm) and the top one (180 cm). The average concentration over the 5 sampling heights is also monitored during injection of CO2.

Leakage 240
Leakage describes the loss of pollutants and air due to overpressure, through the corners and door of the chamber. The leak rate is quantified experimentally using 2 complementary methods. The first method, called "pressure method", uses injection of air inside the chamber. A constant flow of air is injected at a precise value.
Simultaneously, pressure is monitored, and when it gets stable, it means that the flow or air exiting the chamber (e.g. leak flow) is the same as the one injected. This experiment is realized with different flow rates from 245 0.1 L.min -1 up to 6.0 L.min -1 .
The second method uses measurement of the CO2 concentration, and is referred to as the "CO2 method". High concentrations of CO2 are injected inside the chamber, and the concentration decay is observed at constant https://doi.org/10.5194/amt-2021-43 Preprint. Discussion started: 8 April 2021 c Author(s) 2021. CC BY 4.0 License.
pressure. This method was used by Papapostolou et al. (2011) with the CO decay. The CO2 concentration decreases exponentially due to dilution by compensation of air injected into the chamber, and to leakage at a 250 given relative pressure varying between 0.17 Pa and 18.42 Pa. The decay coefficient is converted into a leak flow, expressed in L.min -1 , thus allowing comparison with the "pressure-method" results. The leak flow (Fleak) can also be expressed as a leak rate (Rateleak), given as the percentage of the total volume exiting the chamber in 1 hour, according to Eq. (1).

Wall losses 255
The leakage and size-dependent wall loss correction method presented in this study is based on 4 consecutive steps.
Step 1 consists in correcting total [PM] using the decay of the black carbon concentration [BC]. During step 2, the particle distribution is corrected using a size-dependent wall loss coefficient. This coefficient is based on the theory of Crump and Seinfeld (1981), with an arbitrary estimation of the turbulence.
Step 3 consists in optimizing the turbulence parameter ke to fit corrected data with results of step 1. Finally, step 4 is the 260 computation of the total particle number and mass corrected concentrations.

• Step 1: [PM] correction using [BC]
The first step consists in correcting the total particle mass, using BC as a tracer for primary particle emissions, as done by Grieshop et al. (2009). As BC is an inert compound (Platt et al., 2013;Wang et al., 2018a), the decay of its concentration is due to leakage and wall deposition, with a loss rate kBC. By assuming that particles are 265 internally mixed, total particle mass has the same loss rate as BC (Grieshop et al., 2009;Hennigan et al., 2011;Platt et al., 2013). A corrected PM concentration can thus be obtained at all times of each experiment (Grieshop et al., 2009;Hennigan et al., 2011;Platt et al., 2013) with Eq. (2).
For some experiments, the decay of [BC] cannot accurately be fitted by a 1 st order exponential decay. In those

cases, the term exp(kBC t) is replaced by [BC]t0/[BC]t. 270
This corrected concentration [ ] 1 ( ) represents the evolution of particle mass due to nucleation or condensation/evaporation of organic material. Particles lost to the walls are assumed to be at equilibrium with suspended ones (Grieshop et al., 2009).

• Step 2: Particle-size-dependent correction
The second step consists in correcting particle mass and number concentration, accounting for the size-275 dependence of wall deposition (Charan et al., 2018;Leskinen et al., 2015;Pierce et al., 2008;Wang et al., 2018bWang et al., , 2018a. The aerosol wall deposition rate due to turbulent diffusion, Brownian diffusion and gravitational sedimentation is given in Eq. (3) (Corner and Pendlebury, 1951;Crump and Seinfeld, 1981). It is given for a cubic chamber of side length L, as a function of particle diameter Dp. https://doi.org/10.5194/amt-2021-43 Preprint. Discussion started: 8 April 2021 c Author(s) 2021. CC BY 4.0 License.
With: 280 i the SMPS particle size channel of geometric midpoint diameter Dp ke the eddy diffusivity coefficient (sec -1 ) All terms of this equation, except for ke, can be found using literature (Crump and Seinfeld, 1981;Seinfeld and Pandis, 2016) and experimental data. They are listed in Appendix B. The ke coefficient represents the turbulence inside the chamber, induced by the exhaust gas injection, the air flow to compensate leakage and sampling, and electrostatic forces near the walls (Charan et al., 2018;Crump and Seinfeld, 1981;Pierce et al., 2008;Seinfeld and Pandis, 2016;Wang et al., 2018a). As it cannot be measured (Charan et al., 2018), ke will in a first instance 290 be given an arbitrary value in order to compute βi (Dp). The rate of leakage is taken as defined by Schnell et al. (2006) as the ratio of the flows entering (or exiting) the chamber V̇ by the volume of the chamber V: = V̇V ⁄ . It follows similar basis to the "pressure method" defined above. It assumes that at constant pressure, the flow of air injected into the chamber is equal to the sum of leakage and of what is sampled by the instruments. It is determined for each experiment, using the flow of air injected into the chamber V̇ (L.sec -1 ) and the total volume 295 of the chamber V. The loss of particles due to leakage and wall deposition in each size bin can then be estimated for the arbitrary value of ke. Assuming that the coefficients α and βi (Dp) are constant over the course of each experiment, the loss process is a first-order exponential decay (Leskinen et al., 2015;Nah et al., 2017;Pierce et al., 2008;Verheggen and Mozurkewich, 2006;Wang et al., 2018aWang et al., , 2018bWang et al., , 2014 given by Eq. (4).
By adding this to the measured distribution at time t ([ ] ( )), a distribution corrected for leakage and 300 wall deposition can be obtained. It represents the particle mass distribution evolution, solely due to photochemical and physical processes (Eq. (5)).
The corresponding total mass concentration can be computed with Eq. So far, the evolution of the total particle mass corrected for leakage and wall deposition has 2 different expressions. They both represent the evolution of total particle mass in the chamber solely due to photochemical and physical processes of nucleation and condensation (coagulation can also occur, but has no effect on particle mass). However, the expression obtained during step 2 ( [ ] 2 ( )| ) is attached to high uncertainties as it is found for an arbitrary value of ke. Charan et al. (2018) realized an optimal fitting of 310 experimental data to determine the parameter ke. Following the same idea, an optimization of ke is performed to minimize the difference between the corrected PM concentrations obtained during step 1 (considered as the reference) and during step 2 (Eq. (7)).
This operation results in an optimized expression of βi (Dp), for each chamber experiment. It is used in the equations of step 2 (Eq. (3), Eq. (4), Eq. (5), Eq. (6)), to obtain corrected distributions and total concentrations. 315 The values of βi (Dp) are used to compute a mean wall loss coefficient βke mean , depending on the ke value found after optimization. To account for the SMPS bins, each βi (Dp) is weighted with the relative size of the channel i (Dp), according to Eq. (8). With: -Dp,i the diameter associated with the bin i, with Dp ranging from 14.1 nm to 615.3 nm 320 -Dp,i+1-Dp,i the size of the bin i, where Dp,max+1=637.8 nm -∑( , +1 − , ) = 623.7 the total measurement range considered in this study The optimized coefficient α+βke mean represents the rate of particle losses due to leakage and size-dependent wall deposition.

• Step 4: Corrected number distribution and number concentration 325
The size-dependent optimized loss coefficient α+βi (Dp) can be applied to the number distribution to determine what is lost due to leakage and wall deposition (Eq. (9)).
This lost distribution can be added to the measured one, to obtain a corrected number distribution (Eq. (10)). The corresponding corrected total number concentration can be computed (TSI, 2010) with Eq. (11). This method can also be used to determine the size-dependent loss rate βi (Dp) for seed only experiments, with 330 ammonium sulphate particles. The evolution of those particles is not impacted by effects of condensation or evaporation. This means that the corrected total mass of ammonium sulphate particles should be a constant, equal to the concentration at time t0, as expressed in Eq. (12). This corrected mass in used as the reference in Eq. (7) of step 3. The mass concentration decay is associated to a rate kPM amm sul , which represents the losses due to leakage and wall deposition only. It can therefore be compared to the rate kBC found for exhaust particle 335

Mixing time
Mixing time is investigated using CO2 measurements at 3 sampling heights as presented in Figure 1a. They show 340 that just 1 min after the end of the CO2 injection, the vertical gradient of CO2 is important. The mean CO2 concentration is 0.977 ± 0.013 %, but this value reaches 0.964 % at the top of the chamber and 0.990 % at the bottom. After 20 minutes, the mean concentration is 0.972 ± 0.003 %, with a minimum value of 0.969 % at the top and 0.975 % at the bottom. From that time, the concentration variability is less than 1 % between the minimum and maximum values, and the mixture can be considered homogeneous. After 1 hour, the mixture is 345 homogeneous if the instrument uncertainties are accounted for.

350
The CO2 concentration averaged over the 5 sampling heights is shown in Figure 1b. Results show that the mean concentration increases rapidly during the first 20 minutes following injection. Then it reaches its maximum and remains stable. At this stage, the mixture can be considered as homogeneous. The results of the CO2 tests indicate that the mixture inside the chamber can be considered as homogeneous 20 minutes after the end of injection. Therefore, in this study, all the chamber analysis will start 20 minutes after the 355 end of the exhaust gas injection.

Leakage
The leak flows obtained with the "pressure method" and the "CO2 method" are presented in Figure 2 with the green and black curves respectively. Both methods give similar results. A numerical fit of the curves gives a mathematical expression (Eq. (13)) for the leak flow Fleak (L.min -1 ) as a function of the relative pressure ΔP (Pa) between the chamber and outside. 365 For values of relative pressure between 0.2 and 5 Pa, the leak flow is between 0.4 and 2.9 L.min -1 , corresponding to a leak rate between 0.3 and 2.2 %vol.h -1 . Platt et al. (2013) reported for the PSI chamber an average leak rate of 0.08 %vol.h -1 . This is lower than the one found in this study, probably due to fact that the chamber is operated under overpressure conditions. This leak rate can be used to estimate gas and particle phase leakage from the measured relative pressure. For this study, a leak rate is estimated for each experiment, without the dependency 370 on the relative pressure. Indeed, it can change between the end of injection and the remaining of the chamber experiment (the leakage due to high overpressure measured right after injection is not accounted for, but decreases rapidly). Another leak definition is given in the wall losses section, with similar basis as the "pressure method", and taking into account the instruments sampling flows. https://doi.org/10.5194/amt-2021-43 Preprint. Discussion started: 8 April 2021 c Author(s) 2021. CC BY 4.0 License.

Wall losses 375
The 4-step correction method is applied to more than 50 experiments using particles from passenger car exhausts (diesel or gasoline) and ammonium sulphate particles. The associated experimental protocols are described in the part Methods. The optimization processes give ke values between 0.001 and 27.32 sec -1 , with an average of 2.41 sec -1 . The corresponding wall loss coefficients are respectively 1.05×10 -4 , 1.58×10 -2 and 4.69×10 -3 min -1 for a particle of diameter 100 nm. These values are similar to those reported by Leskinen et al. (2015) and Babar et 380 al. (2017), of 7.50×10 -4 and 3.96×10 -3 min -1 , respectively. Figure 3 shows the size distribution of the wall loss coefficient βi (Dp) as well as the corresponding mean wall loss coefficient βke mean for 3 values of ke in this range: ke=0.36 sec -1 ; ke=2.57 sec -1 ; ke=16.70 sec -1 . These values are chosen to represent the lower part of the range, the average, and the upper part of the range, respectively. The mean wall loss coefficients are respectively 1.31×10 -3 min -1 ; 8.44×10 -3 min -1 ; and 2.13×10 -2 min -1 . 385  Figure 3 shows that for a given ke, wall losses vary greatly as a function of the particle diameter in the range 15-600 nm. They are higher for smaller particles, due to higher Brownian diffusion. These results also show that 390 the wall loss coefficients are very sensitive to the parameter ke. Differences increase as the particle diameters decrease. The value of the eddy diffusivity ke depends on the turbulence induced by the injection into the chamber and the electrostaticity of the Teflon walls, which can change between 2 experiments. Since experimental conditions greatly influence the turbulence and the wall deposition (Nah et al., 2017;Wang et al., 2018a), it is important to account for experiment-to-experiment variations when correcting particle losses. 395 In about a third of the chamber experiments, total particle losses cannot be approximated by a simple exponential decay of [BC]. These experiments usually took place shortly after the chamber walls were manually cleaned.
These actions resulted in an important electrostatic charge induced to the walls, thus increasing particle losses.
For these "charged walls" experiments, the optimized values of ke range from 0.62 to 27.32 sec -1 , with an average of 7.98 sec -1 . These values are high but still lower than some values (36 and 269.4 sec -1 ) found in 400 literature (Crump and Seinfeld, 1981;Okuyama et al., 1986). and 0.83×10 -3 min -1 . The mean α+βke mean values for the "neutralized walls" experiments are 3.4 (diesel) and 1.7 (gasoline) times lower than for the "charged walls" experiments. This is close to what was found by Wang et al.
(2018a), with particle loss rates 3-4 times higher between "undisturbed" and "disturbed" experiments.  when the walls were electrostatically charged. The first experiments performed with charged walls (e.g. with the 430 highest electrostatic charge) were with diesel engine particles. This explains the very high values reached by the diesel "charged walls" coefficients. Ammonium sulphate particles have lower loss coefficients than engine particles, and remain in a narrow range. This could be explained by their nature and chemical composition. Also, the flow at which they are introduced in the chamber (5.6 L.m -1 ) is much lower than for vehicle exhaust experiments (20-60 L.min -1 ). The turbulence and Brownian diffusion are therefore lower, and wall deposition 435 decrease. For the ammonium sulphate experiments, the loss coefficient (leak + wall losses) found with the 4-step correction method range from 8×10 -6 sec -1 to 2×10 -5 sec -1 , for particles of diameter 100 nm. Nah et al. (2017) studied the wall loss coefficient ammonium sulphate of particles in a 12 m 3 chamber. They found values of β at 100 nm around 1-5×10 -5 sec -1 . The values found in this study are in pretty good agreement with what was found by Nah et al. (2017). This indicates that the correction method developed in this study gives good results, 440 consistent with what is found for other chambers of comparable size. Figure 4 shows that the nature of the particles is likely to impact the loss coefficients. This shows the advantage of the 4-step correction method, which isn't based on the assumption that the particles of the study have the same losses as ammonium sulphate

Figure 4: Mean loss coefficient α+βke mean (accounting for channel size) for the chamber experiments performed with particles from diesel (light grey) and gasoline (white) engines, as well as ammonium sulphate particles (dark grey). Coefficients for the diesel and gasoline experiments are sorted in 2 categories: "neutralized walls" experiments for which [BC] evolution follows a simple exponential decay (dotted boxplots) or "charged walls" experiments for which [BC] evolutions follows a double exponential decay (hatched boxplots). The boxes represent the 25 th and 75 th
particles. Figure 4 also shows that the electrostatic state of the walls appears to be the most important factor determining particle losses to the walls. 445 The coefficients α+βke mean and kBC both represent the rate at which particles are lost due to leakage and wall deposition, and should in theory be correlated. To investigate this, α+βke mean coefficients of "neutralized walls" experiments (e.g. for which [BC] follows a simple exponential decay giving a kBC coefficient) are plotted in   Figure 5 also show that loss rates of ammonium sulphate particles are generally lower than those from diesel or gasoline experiments, as found previously (Figure 4). Even though wall charge appears as the dominant factor for particle wall losses, the nature of the particles also seems to have an impact. This shows the interest of using the 4-step 460 correction method described above.

Emission factors of particles, BC and IVOCs from diesel vehicles
Vehicle emissions are quantified in order to discuss the impact of each vehicle type and driving condition on    This is close to what was reported by Zhao et al. (2015), with a ratio 7 in creep conditions between cat DPFequipped and nonaftertreatment vehicles. Moreover, the driving conditions also impact IVOC emissions. For the DPF-equipped vehicles, IVOC EFs are 1.4 to 6.2 times higher in UC conditions than in MW conditions. This is 500 due to more complete combustion at high temperature operations. This follows the same trend as what was reported by Zhao et al. (2015) for diesel vehicles, with ratios from 6 to 23 between creep/idle and high speed conditions. For the Euro 3 (D1) and the Euro 4 (D2) vehicles, IVOC emissions are respectively 1.2 and 1.8 times higher in UC conditions. This could be due to the cold start.
Moreover, for non-DPF vehicles (D1 and D2), IVOC emissions are dominated by n-and b-alkanes, which 505 account for 64 %. Naphthalene accounts for 6 % of IVOC emissions. Cycloalkanes and aromatics account respectively for 14 and 16 % of IVOC emissions. These results are consistent with those obtained by Lu et al. (2018) and Zhao et al. (2015). They found IVOC emissions of non-DPF diesel vehicles to be dominated by cyclic compounds and alkanes (mainly unspeciated), with also a fraction of aromatics. For the cat DPF vehicle (D5), alkanes represent 78 % of IVOCs, followed by cycloalkanes (19 %) and then aromatics and naphthalene (2 510 and 1 % respectively). For the vehicles with add DPFs (D3 and D6), only alkanes are identified in the IVOC retention time range.
(5.6 ± 3.1) % of the total IVOC mass given by Zhao et al. (2015) was identified. This brings the IVOC/NMHC 520 ratios to (4.9 ± 2.2) % (nonaftertreatment) and (8.4 ± 9.1) % (DPF-equipped), if considering only the identified portion of IVOCs. This is higher than what is found in this study, but remains in the same range if uncertainties are accounted for.
Certain n-and b-alkanes corresponding to SVOCs are also sampled on the sorbent tubes. Their emissions are higher for the Euro 3 (D1), Euro 4 (D2) and cat DPF Euro 5 (D5) vehicles, with average EFs of 1093 and 525 1520 µg.km -1 in UC and MW conditions respectively. For the Euro 5 (D3) and Euro 6 (D6) vehicles, both equipped with add DPFs, the average SVOC EFs are 263 and 247 µg.km -1 in UC and MW conditions respectively. However, their quantification is performed with response factors from compounds with different carbon numbers (C20 for vehicles D1, D5 and C16 for vehicles D2, D3, D6). This can induce important uncertainties in their EFs. 530 Considering the French fleet (André M. et al., 2014) and EFs of Figure 6, the contribution of each Euro norm to particle and IVOC emissions from diesel passenger cars is estimated. Since no vehicle of norms pre-Euro, Euro 1 and Euro 2 was tested, their emissions are assumed to be in the same range as those from Euro 3 vehicles. They are grouped in a category named Euro pre-1-2-3. Figure 7a gives the French passenger car fleets from 2015 to 2030 (André M. et al., 2014). Figure 7b and Figure 7c give the evolution of total PM and IVOC emissions, 535 computed as the product of emission factors and fleet composition (assuming a constant number of vehicles).  For IVOCs, emissions are also due to DPF-equipped vehicles (Euro 5 and Euro 6). The share of those vehicles in IVOC emissions by diesel PCs is less than 1 % in 2010, 46 % in 2020 and 78 % in 2030. More modern vehicles will be the main contributor to IVOC emissions in 2030, with effects on particle physical and photochemical evolutions. More investigations should therefore be conducted to estimate their emissions and evolutions. 555

Evolution of particles from diesel exhaust
In this part, the physical processes having effects on particle number, mass and size are investigated, for the 6 diesel vehicles (Euro 3 to Euro 6) in both UC and MW conditions. Data are corrected for leakage and wall losses, using the new 4-step correction method. Figure 8a gives the hourly increase of [PM], for the vehicles classified by Euro norm. It shows that the Euro 3 and Euro 4 vehicles have the higher [PM] hourly increases. 560 Increases are at least of 5 %.h -1 during more than 500 min, and up to 17 %.h -1 during 100 minutes for the Euro 3.
The Euro 4 vehicle has hourly [PM] increases in the same range, and undergoes an increase during an average of 400 minutes. This increase in [PM] can be explained by condensation of organic material onto preexisting particles. The emissions of IVOCs observed in Figure 6 for the Euro 3 and Euro 4 vehicles seem to confirm this explanation. The IVOC emissions from the Euro 5 vehicles could explain the slight increase in particle mass 565 during evolution, presented in Figure 8a. This increase for the Euro 5 vehicles can go up to 5 %.h -1 during almost 5 hours. The Euro 6 vehicle however, does not undergo any [PM] increase, which is consistent with the very low emissions of precursors.   Figure   9a. It is plotted as a function of the initial particle surface, and the labels give the hourly [PM] increase.  Figure 9 clearly shows that when the available particle surface is high (> 10 -3 cm 2 .cm -3 ), the time to reach the maximum [PM] is quite short, and doesn't exceed 2.5 hours. However, for initial surfaces below 4×10 -4 cm 2 .cm -3 , condensation seems to be a very slow process. Even though uncertainties become large after 6 600 hours of evolution, it seems that condensation can occur slowly during 7 to 10 hours. This difference between fast and slow [PM] increases is shown in Figure 9b  occurring during 105 minutes. This fast condensation seems to be mostly determined by the high particle surface available.
Other than condensation, nucleation and coagulation are 2 processes that can have important effects on particle concentrations and distributions. Both of those processes are investigated, and results are shown in Figure 10.   Figure 10c and Figure 10d give the evolution of the particles in 3 diameter bins, for experiments with initial [PN] of 1.3×10 5 #.cm -3 and 2.5×10 5 #.cm -3 respectively. In Figure 10c, the concentration of the particles in the bin 46.1 nm decrease rapidly during 20 to 30 minutes. Simultaneously, the concentration of larger particles (in the 625 bin 250.3 nm) increases. This seems to indicate that particles around 46.1 nm coagulate to form larger particles.
The smaller particles (around 18.1 nm) have a constant concentration. Assuming that those particles should also undergo the process of coagulation, their constant concentration could be explained by the process of nucleation.
It would compensate the loss due to coagulation. It could also be compensated by evaporation of organics from larger particles. This last explanation is however unlikely since total mass tends to increase and concentrations of 630 larger particles also increase. The same trends are observed in Figure 10d  effect, increasing particle number. However, nucleation alone is not likely to occur during 2 hours. This increase 655 can also be explained by the growth of small particles (formed at the beginning of the evolution) that became detectable in the SMPS range. Figure 11f, Figure 11g and Figure 11h show relative [PN] evolution with initial [PN] in the range [1-8]×10 4 #.cm -3 , and show medium to high particle increase. When initial particle concentration goes over [8-9]×10 4 #.cm -3 , the main evolution becomes negative. It remains negative during 40 to 120 minutes. It can reach -27 %.h -1 , clearly showing that coagulation is occurring. Figure 11c, Figure 11d and 660 Figure 11e give examples of the [PN] decrease for those high PN initial concentration conditions. There seems to be a breakthrough around [8-9]×10 4 #.cm -3 , where the evolution trend changes. The equivalent surface concentrations are [1.2-2]×10 -4 cm 2 .cm -3 . Under this threshold, the particle number increase prevails, partially due to nucleation. This is consistent with Ning and Sioutas (2010), who show that organic vapors nucleate more easily when particle concentrations are low, and condense more easily when particle 665 concentrations are high. This is also consistent with Figure 8, which shows that high initial [PN] leads to higher hourly [PM] increases. Above this threshold, particle number decreases over time, and coagulation becomes important, and the most significant process affecting particle number.
These results are in good agreement with tunnel measurements performed by Imhof et al. (2006). They showed that nucleation would occur more easily under low traffic density (e.g. with lower initial particle concentrations). 670 They also showed that PN nucleation mode concentrations decrease when soot mode particle concentrations are high. Moreover, they found that nucleation mode particles increase until the surface area of soot particles reaches a threshold of 5×10 -5 cm 2 .cm -3 . This threshold found experimentally in 2 tunnels is close (factor 2-4) to the one found in this study ([1.2-2]×10 -4 cm 2 .cm -3 ) in laboratory conditions. Finally, Imhof et al. (2006) found similar trends of traffic influence on particle evolutions as in this study. Condensation is more likely to occur in the 675 presence of highly particle-emitting vehicles. Nucleation may be more important when traffic is composed of less polluting vehicles. Results of this study show that the processes affecting particle evolutions is greatly influenced by atmospheric concentrations, which is dependent on traffic density and fleet composition.

Conclusion
This study presents the characterization of a new 8 m 3 environmental chamber with Teflon walls, meant to study 680 the physical evolution of primary pollutants emitted by road traffic. A new size-dependent method to correct particle losses due to leakage and wall deposition was developed and applied. It accounts for experiment-toexperiment variations. It consists in 4 steps, using the [BC] decay and wall loss coefficients from the theory of Crump and Seinfeld (1981). These 2 complementary parts are in good agreement. The total loss coefficients α+βke mean of "charged walls" experiments are in the range [4.19-11.05]×10 -3 min -1 for diesel and 685 [2.04-5.01]×10 -3 min -1 for gasoline. For "neutral walls" experiments, they are in the range [1.25-4.09]×10 -3 min -1 for diesel, [0.93-3.93]×10 -3 min -1 for gasoline and [0.63-1.01]×10 -3 min -1 for ammonium sulphate. The wall charges appear to be the most important factor affecting particle wall losses. Results of wall losses obtained from ammonium sulphate particle experiments show similar trends as those found in the literature for a chamber of comparable size. 690 EFs of PN, PM, BC and IVOCs for Euro 3 to Euro 6 diesel vehicles were studied in detail in order to understand their role on particle evolution in dark conditions. For non-DPF vehicles (D1 and D2), IVOC emissions are https://doi.org/10.5194/amt-2021-43 Preprint. Discussion started: 8 April 2021 c Author(s) 2021. CC BY 4.0 License. dominated by n-and b-alkanes, (64 %), followed by cycloalkanes (14 %) and aromatics (16 %). For the cat DPF vehicle (D5), alkanes represent 78 % of IVOCs, followed by cycloalkanes (19 %) and then aromatics (2 %). For the vehicles with add DPFs (D3 and D6), only alkanes were identified in the IVOCs. 695 Moreover, this study presents results of evolution in the dark of particles emitted by diesel passenger cars. It is shown that [PM] increase can reach 17 %.h -1 , without the effects of photochemistry.
[PM] increase appear to be correlated with initial [PN]. The [PM] increase of older diesel vehicles (Euro 3 and Euro 4) is enhanced in comparison to modern vehicles. This is due to their higher emissions of BC, PN and PM, compared to DPF-equipped vehicles, as well as their emissions of IVOCs. Also, the condensation process is shown to be 700 highly dependent on the available particle surface. Condensation is 4 times faster when the available surface is multiplied by 3. Finally, this study shows that the [PN] decrease (by coagulation) or [PN] increase (by nucleation or condensation on particles below 14 nm) is dependent on the particle concentration. For initial [PN] below [8-9]×10 4 #.cm -3 , nucleation or condensation on particles below 14 nm are probably the main processes affecting particle number. Above this threshold, coagulation prevails, due to higher probability for particles to collide. 705 This is also explained by the fact that at high PN concentrations, gas-phase organics are more likely to condense onto preexisting particles than to nucleate.
Results found in this study under laboratory conditions are in good agreement with tunnel observations described in the literature. They can help better understand the conditions under which physical processes are more likely to occur. They can be applied to several conditions in the dark, such as winter rush hours or tunnel evolutions. 710 Data availability. All data from this study are available from the authors upon request. Competing interests. The authors declare that they have no conflict of interest.

Acknowledgements. 725
Financial support. This work was supported by the ADEME CORTEA program with the project CAPVEREA and MAESTRO and the ANR program with the project POLEMICS.