As mentioned previously, CRM is a technique based on monitoring competing reactions of OH with a reference compound (pyrrole) and ambient trace gases. Pyrrole was initially chosen as the reference compound for CRM by Sinha et al. (2008b) for several reasons. First, the only known atmospheric source of pyrrole is biomass burning (Karl et al., 2007), and due to its high OH reactivity, ambient concentrations of pyrrole are usually negligible. The rate constant for the reaction of pyrrole with OH (1.28×10-10 cm3molecule-1s-1) (Dillon et al., 2012) has also been well-established and is sufficiently fast to compete with the reactions of other trace gases with OH. Finally, pyrrole can easily be detected by mass spectrometric methods.

A schematic of the MD-CRM instrument is shown as Fig. 1. The instrument consists of a glass reactor that is supplied with various gases (zero air, high purity nitrogen, pyrrole, standard gases) and interfaced to a mass spectrometer, as shown in Fig. 1a. Gas flows are directed through the instrument with solenoid valves (Burkert 6606) and mass flow controllers (MKS 1179A), which are driven by an analog/digital conversion board (National Instruments NI PCI-6024E). The OH reactivity measurements are fully automated and coordinated by a software program written within the LabVIEW environment (National Instruments).

The glass reactor is based on the design proposed by Sinha et al. (2008b) and features multiple arms (numbered 1 through 4 in Fig. 1b), through which all the required gases are introduced. Approximate flow rates through all four arms are indicated in Fig. 1b. Pressure and temperature within the CRM reactor are recorded by a pressure gauge (MKS Baratron) and a thermocouple attached to arm 3. To produce OH, humidified nitrogen (Praxair, 6.0 purity) is passed through an ultraviolet mercury pen lamp (UVP Pen-Ray, 11sc1) located in arm 2. Water vapor is photolyzed by the 184.9 nm emission of the lamp to produce OH via Reaction (R1). H2O+hνλ≈184.9nm⟶H⚫+OH⚫ However, this source of OH also leads to the formation of HO2, since H atoms produced in Reaction (R1) can rapidly react with oxygen at the exit of arm 2 via Reaction (R2). H⚫+O2⟶MHO2⚫. OH can also be produced by the photolysis of ozone in the CRM reactor and the subsequent reaction of excited oxygen atoms with water vapor.

Humid zero air is produced by sampling ambient air through a catalytic converter, which consists of a stainless steel tube (1.27 cm diameter, 20 cm length) filled with 2 g of platinum wool (Shimadzu) and heated to a temperature of 350 ∘C. The platinum wool leads to the complete oxidation of ambient VOCs to water and carbon dioxide. The zero air humidity is therefore close to ambient relative humidity (RH). The air passes through a 2.6 m of tubing before entering the CRM reactor; this should ensure that the air going through the catalytic converter has been thermalized. While there is the possibility of a slight temperature change between the C2 and C3 conditions (see Sect. 2.2.2), the mercury lamp used to generate OH also releases some heat through arm 1 and tends to buffer the reactor temperature between the C2 and C3 conditions. Measurements of temperature within the CRM reactor (ranging from 24 to 30 ∘C) suggest that any increase in temperature compared to ambient is lower than 10 ∘C. Over this range of temperatures, the rate constant for OH + pyrrole differs by less than 5 % based on the temperature-dependent relationship derived by Dillon et al. (2012); this is within the measurement uncertainty for the CRM instrument. The catalytic converter was tested with mixtures made of several tens of hydrocarbons (a few ppbv of each) and was found to efficiently remove VOCs to levels lower than 5–10 pptv (i.e., the detection limits of GC analyzers). During ambient measurements, ambient air is pulled through a 3 m long sampling inlet made of Teflon tubing (0.64 cm o.d.) at a rate of 415 cm-3min-1 (Fig. 1b). This air (zero or ambient) is mixed with pyrrole and introduced into the reactor through arm 1. A small flow of a calibration gas (up to 50 cm3min-1 of ethane, 50 ppmv, Praxair) can be added in the sampling line to track the performance of the CRM system during ambient measurements.

A second-generation Proton Transfer Reaction-Time-of-Flight Mass Spectrometer (PTR-ToFMS) (Kore Technology, Ely, UK) samples from the CRM reactor through arm 4. The PTR-ToFMS is used as a detector of pyrrole; the protonated form of pyrrole is detected over an m/z range of 68.0±0.4 Dalton. During the intercomparison, the PTR-ToFMS was operated at a reactor temperature of 40 ∘C and an E / N ratio of 136 Td.

As highlighted by Sinha et al. (2009), the sensitivity of PTR-MS instruments toward pyrrole is dependent on humidity, and the pyrrole signal must be carefully calibrated for relative humidity changes within the CRM reactor. The approach described by de Gouw and Warneke (2007) was employed in this study to account for the effect of water vapor on the PTR-ToFMS sensitivity. This approach involves a normalization of the pyrrole signal to a sum of reagent ion signals (H3O++XR× H3O+ ⚫ H2O) that leads to a normalized signal for pyrrole that is independent of humidity. XR, a scaling factor for the H3O+ ⚫ H2O signal, is determined experimentally by measuring the pyrrole signal from a standard mixture under different humidity conditions. For the MD-CRM instrument, XR was determined to be 0.6 (Fig. S2).

The CRM measurement sequence is described in detail by Sinha et al. (2008b) and summarized below. There are three distinct steps in this sequence with the pen lamp on (referred as C1, C2, and C3 below). Another step (C0) is defined as the pyrrole concentration inside the CRM reactor when the mercury pen lamp is off. Each step represents a particular set of operating conditions, which in turn yields a corresponding steady concentration of pyrrole within the reactor (Fig. S3).

During the first (C1) step, pyrrole and dry zero air are introduced into the reactor through arm 1 together with dry nitrogen through arm 2. In the absence of water, OH radicals are not produced, and a baseline C1 concentration is recorded. The relative difference between C0 (lamp off) and C1 (lamp on), i.e. (C0–C1) / C0, is used to estimate the fraction of pyrrole that is lost from direct photolysis inside the reactor, due to photons leaking through arm 2. For the version of the MD-CRM instrument used in this study, approximately 25 % of pyrrole is lost when the lamp is turned on; this is discussed further below.

During a second (C2) step, dry zero air (arm 1) and dry nitrogen (arm 2) are replaced by humid zero air (same RH as in ambient air) and humid nitrogen, respectively. These wet conditions lead to the formation of OH and HO2 radicals through Reactions (R1) and (R2) as discussed above. Pyrrole reacts with OH, titrating it in milliseconds, which results in the lowering of the pyrrole concentration in the reactor to C2. The difference between C1 and C2 provides information on the amount of OH radicals that react with pyrrole. An apparent pyrrole-to-OH ratio can be calculated from C1 / (C1–C2). For the MD-CRM instrument, this ratio ranges from 1.5 to 1.8; this is discussed further in Sect. 2.2.3.

During the third (C3) step, humid zero air flowing into arm 1 is replaced by ambient air, keeping the same flow of humid nitrogen in arm 2. Trace gases present in ambient air compete with pyrrole for the available OH radicals. The decrease of pyrrole observed between step 1 and 2 (C1–C2) is offset by the reaction of OH with trace species in the ambient air, and the pyrrole concentration increases from C2 to C3. The magnitude of this increase depends on the amount of OH reactivity in ambient air.

Sinha et al. (2008b) showed that based on pseudo first-order assumptions (i.e., [pyrrole] ≫ [OH]), the total OH reactivity kOH can be calculated from measurements of C1, C2, and C3: kOH=(C3-C2)(C1-C3)kpC1, where kp represents the bimolecular rate constant for the reaction of OH with pyrrole (1.28×10-10 cm3molecule-1s-1) (Dillon et al., 2012).

The measurement scheme used during this study for ambient OH reactivity measurements is described below and summarized in Fig. S3. Measurements of C0 were performed approximately once per day by manually turning off the mercury lamp during the C1 (dry conditions) and C2 or C3 (wet conditions) measurement steps. The measurements of C0 were used to calculate the sensitivity factor of the PTR-MS instrument toward pyrrole (denoted as RF) under humid and dry conditions, as shown in Fig. S4. Measurements of C0 were performed 11 times during the intercomparison period and were used to track the sensitivity of the PTR-MS and the stability of the flow rates over the course of the intercomparison. This figure shows that the stability of the sensitivity was better than 4 %.

C1 was measured every 12 h for a duration of 2 h. As suggested by Sinha et al. (2008b), dry air was used to measure C1 during the intercomparison. The length of this step is necessary to reach the driest conditions possible to minimize residual OH in the reactor, as shown by a plot of the pyrrole signal over time during a C1 measurement (Fig. S5). However, this procedure could result in an underestimation of C1, as it is difficult to remove all trace amounts of water in the reactor. An alternative method to determine C1, which involves the addition of a VOC to scavenge OH in the reactor, has been proposed by Zannoni et al. (2015). The use of this method after this intercomparison study showed that C1 measurements made using dry air in the MD-CRM reactor were higher by about 10 % than those determined using the OH scavenger (Michoud et al., 2015; Zannoni et al., 2015). It is likely that the difference can be attributed to OH radicals produced by photolysis of residual humidity. For this study, the loss of pyrrole under near-dry conditions when the mercury lamp is turned on was determined to be constant (Fig. S9) throughout the campaign at approximately 25 %. However, this value of 25 % likely represents an upper limit of pyrrole photolysis for this version of the CRM instrument. The impact of an underestimation of C1 on OH reactivity measurements is discussed in Sect. 3.2.1.

After a C1 measurement is completed, C2 and C3 are measured alternately every 2.5 min. The time interval of 2.5 min consists of two subintervals. First, there is 30 s of flushing/equilibration time after a change in the sampling conditions, which ensures that pyrrole has reached a steady concentration in the reactor. After the equilibration is complete, 2 min of pyrrole signal measurements are made at a time resolution of 10 s. An automated calibration test is carried out every 3 h by adding 50 cm3min-1 of ethane (50 ppmv, Praxair) during a C3 measurement. The additional OH reactivity due to the standard addition is determined by subtracting an average value of ambient OH reactivity from the two surrounding measurements made before and after the standard addition. This test is further discussed in Sect. 3.2.1 and allows instrumental performance to be tracked during the campaign.

Several tests were carried out to characterize the MD-CRM instrument, both in the laboratory before the intercomparison and during the field measurements. These tests are necessary to correct known measurement artifacts associated with the CRM technique and are discussed below.

Sinha et al. (2008b) and Dolgorouky et al. (2012) showed that a variation of humidity between the C2 (wet zero air) and C3 (ambient air) measurements could lead to a change in OH production in the CRM reactor, which in turn could result in a C2 measurement that is not representative of the OH production rate observed during the C3 measurement. For example, as relative humidity increases, the OH concentration increases in the reactor and the pyrrole signal decreases. This artifact is corrected for by first calibrating the C2 sensitivity to humidity and then by correcting C2 to match humidity levels observed during C3.

However, to simplify the derivation of OH reactivity, and to minimize uncertainties, it is desirable to reduce or eliminate the humidity change between C2 and C3. To reduce the humidity correction, ambient air was sampled into a catalytic converter to generate zero air free of VOCs with the same level of humidity as ambient air. Ambient air was first passed through cartridges containing Purafil media (Purafil, Inc.) and activated charcoal located upstream of the catalytic converter to remove NOx species. To test the effectiveness of this setup to remove NOx species, a NOx analyzer was placed at the exit of the Purafil/charcoal trap-catalytic converter system to measure NOx when ambient air was passed through this system. It was found that more than 85 % of ambient NOx was removed. A drawback of this setup was that the adsorption of water onto charcoal altered the relative humidity (RH) of the zero air, leading to a small (up to 10 %) difference in RH between C2 and C3. Because of this, a correction was still necessary.

Tests were conducted throughout the campaign to investigate the C2 sensitivity to humidity by varying RH in the CRM reactor. The absolute water concentration in the CRM reactor was tracked using the m37 / m19 (H3O+ ⚫ H2O / H3O+) ratio calculated from PTR-ToFMS signals. This ratio has been shown to be a suitable metric to track humidity (Inomata et al., 2008) and was found to be proportional to RH for the PTR-ToFMS instrument used in this study. Figure S6 shows that C2 decreases linearly with the m37 / m19 ratio for the range of humidity investigated (≈ 20–60 % RH, 25 ∘C). It must be noted that this dependence is not due to the humidity dependence of the PTR-MS sensitivity toward pyrrole, but is only due to a change in OH production in the reactor, as the normalization procedure described in Sect. 2.2.1 was applied to all pyrrole measurements.

The slope of the linear relationship observed between pyrrole concentration and the m37 / m19 ratio was used to correct the C2 measurements as shown in Eq. (4). C2corr=C2raw+p(m37m19C3-m37m19C2) The value of p used to correct the C2 measurements (C2raw), calculated as an average of the slopes of the two plots shown in Fig. S6 (made on two different days during the campaign), is 0.503 ppbv per percent of change of the m37 / m19 ratio.

As discussed previously, ambient NO can generate a measurement artifact due to the formation of OH in the reactor from the reaction of HO2 with NO. This artifact affects C3, as ambient NO is introduced into the reactor during this step, leading to an underestimation of the pyrrole concentration, since additional OH reacts with pyrrole. In order to determine the correction for C3, the dependence of the pyrrole signal on NO mixing ratios was investigated at an apparent pyrrole-to-OH ratio of 1.7 using zero air alone and zero air mixed with ethane. The HO2+NO artifact was characterized by adding NO at mixing ratios ranging from 5 to 180 ppbv. The decrease in the pyrrole concentration was parameterized as a function of NO, as shown in Fig. S7. This figure shows that the same response is observed for both zero air and zero air + ethane, suggesting that this artifact is independent of ambient OH reactivity. Additional tests performed during laboratory experiments showed that the NO artifact is independent of ambient OH reactivity using compounds other than ethane, including propene and isoprene (Michoud et al., 2015). For NO mixing ratios lower than 50 ppbv, the decrease in C3 is linear with NO and is characterized by a slope aNO of 0.122 ppbv of pyrrole per ppbv of NO and a zero intercept (bNO=0 ppbv of pyrrole). For NO mixing ratios higher than 50 ppbv, a second linear regime is observed, and a different correction is applied using a slope aNO of 0.066 ppbv of pyrrole per ppbv of NO and an intercept bNO of 4.51 ppbv of pyrrole. Similar correction curves were determined for another CRM instrument deployed during the MEGAPOLI field campaign (Dolgorouky et al., 2012). This correction was recently investigated through computer modeling for the MD-CRM instrument and was found to be consistent with our understanding of the CRM chemistry (Michoud et al., 2015). The correction applied on C3 for the impact of NO is calculated using Eq. (5). C3corr=C3+aNO[NO]+bNO In addition to the correction for NO, it was observed that an addition of NO2 into the reactor also leads to a further decrease of the pyrrole signal. This behavior is due to the conversion of NO2 to NO from heterogeneous reactions on surfaces such as stainless steel and photolysis (caused by leakage of photons from the mercury lamp inside the reactor). It must be noted that the degree of conversion of NO2 to NO within the reactor is sensitive to the position of the lamp within the reactor. Because of this, the conversion of NO2 to NO should be characterized for each instrument. The change in the pyrrole signal observed when NO2 is added in the CRM reactor suggests that approximately 46 % of NO2 is converted into NO. A procedure similar to that described for NO above was applied to correct C3 for this additional NO2.

A third correction must be applied to account for a deviation from first-order conditions as discussed by Sinha et al. (2008b). The derivation of total OH reactivity via Eq. (3) assumes that [pyrrole] ≫ [OH] (i.e., pseudo first-order conditions). However, this assumption is not fulfilled within most CRM instruments, as the concentration ratio of pyrrole to OH is usually maintained at a value lower than 2 to enhance the C2–C3 modulation, which in turn improves the detection limit. To correct the CRM measurements for a deviation from pseudo first-order conditions, different concentrations of standards were added inside the reactor to derive a relationship between the CRM measurement and the OH reactivity added in the reactor (Fig. 2). We found that by running the reactor with an apparent pyrrole-to-OH ratio of 1.7, the measured OH reactivity values agree to within 10 % of the OH reactivity expected from the addition of a standard in zero air under both laboratory and field conditions. Furthermore, the correlation between the measured and calculated OH reactivity values seems to be linear, based on the use of three different compounds: ethane (kC3H8+OH=2.4×10-13 cm3molecule-1s-1 at 298 K) (Atkinson et al., 2006), ethene (kC2H4+OH=7.9×10-12 cm3molecule-1s-1 at 298 K and 1 bar pressure) (Atkinson et al., 2006), and propene (kC3H6+OH=2.9×10-11 cm3molecule-1s-1 at 298 K and 1 bar pressure) (Atkinson et al., 2006). The OH rate constants of these compounds span 2 orders of magnitude and are representative of reactive gases found in the atmosphere. The agreement observed in Fig. 2 indicates that the CRM instrument accurately quantifies the OH reactivity. Michoud et al. (2015) provide a detailed characterization of the MD-CRM instrument and show that experimental determinations of this correction agree relatively well with a modeling of the CRM chemistry.

OH reactivity values calculated from Eq. (3), following the C2 and C3 corrections, were multiplied by 0.912 (the slope of Fig. 2) to derive the OH reactivity in the CRM reactor and to take into account the deviation from first-order conditions for an apparent pyrrole-to-OH ratio of 1.7. During ambient measurements, the apparent pyrrole-to-OH ratio changed from 1.5 to 1.8 due to changes in ambient humidity that impact the production of OH in the reactor. However, subsequent characterizations by Michoud et al. (2015) have shown that this correction factor does not change significantly over this range of pyrrole-to-OH ratios. OH reactivity values were finally corrected for each measurement by a factor of approximately 1.25 to account for ambient air dilution based on the sampling (416 cm3min-1 from Fig. 1b) and total (517 cm3min-1 from Fig. 1b) flow rates inside the reactor.

In practice, the limit of detection (LOD) is governed by the minimum detectable difference between C2 and C3. The LOD for the CRM instrument was estimated by determining the standard deviation (σC2) on groups of five successive C2 measurements. An OH reactivity was calculated for each ambient C2 measurement using the corresponding C1 value and assuming a C3 greater than C2 by 3σC2. From the median of these calculated values, it was determined that the MD-CRM instrument has a 3σ LOD of 3.4 s-1. This LOD is comparable to those reported for other CRM instruments, which range from 3 to 6  s-1 (Sinha et al., 2008b, 2010; Dolgorouky et al., 2012; Nölscher et al., 2012b). The measurement precision is discussed in Sect. 3.2.1, based on an assessment of the different corrections that must be applied to ambient measurements.

The UL-FAGE instrument can be used in a quantification mode to measure ambient concentrations of OH and HO2 radicals or in an OH reactivity mode using the pump–probe method described by Sadanaga et al. (2004). For the latter, the FAGE instrument is coupled to a photolysis cell (Fig. 3), in which a plume of OH is generated by laser photolysis of ozone. Time-resolved OH decays are monitored at a resolution time of 200 µs using the high repetition rate probe laser (5 kHz) of the FAGE instrument. Three other instruments, based on the same technique, are used for field measurements (Sadanaga, 2004b; Lou et al., 2010; Stone et al., 2015). A comparison of the specifications of the different instruments can be found in Table 2.

The UL-FAGE instrument has been described previously (Amedro et al., 2011, 2012; Parker et al., 2011), and only a brief description is given below, focusing on the setup used during this study (i.e., with the FAGE inlet connected at the end of the photolysis reactor, Fig. 3). The photolysis cell is a 50 cm long, 5 cm-i.d. cylindrical tube made of aluminum. A Suprasil quartz window is mounted on one side of the cell, and the other side is directly connected to the FAGE nozzle. The pumping from the FAGE instrument ensures that the photolysis cell is continuously flushed with ambient or zero air. Gases are introduced via Swagelok fittings at the quartz window side of the reactor. During the campaign, ambient air was sampled through a 3 m length of 12.8 mm-i.d. Teflon tubing. The pressure in the cell (P∼930 mbar) was lower than atmospheric pressure due to a restriction of the flow through the sampling Teflon line, which was adjusted to 8.3 Lmin-1.

OH is generated inside the cell by ozone photolysis at 266 nm in the presence of water vapor (Reactions R3, R4) using a quadrupled YAG laser (Quantel, YG 981C) operated at a pulse repetition rate of 1 Hz. O3+hν(266nm)⟶O2+O1DO1D+H2O⟶OH+OH.

An ozone mixing ratio of at least 60 ppbv is maintained inside the photolysis cell by injecting a small flow rate of 20 cm3min-1 (negligible compared to the main flow through the reactor) of concentrated ozone using an ozone generator (Scientech). The pulse repetition rate of 1 Hz represents a good compromise between the reaction time (1 s or less), the time needed for air exchange in the cell (6.6 s), and the acquisition time (sum of 30 acquisitions of 1 s). The energy of the photolysis laser was set to 20 mJpulse-1 for a beam diameter of 2.5 cm, which was achieved after expansion through a telescope. This expansion of the beam permits the generation of OH in a cylindrical volume that is larger than the FAGE nozzle (1 mm) in order to probe a more homogeneous volume with respect to the OH concentration, even if the shape of the beam involves a Gaussian distribution. The pulse duration of the photolysis laser is 20 ns (full-width half maximum).

The photolysis cell is coupled with an airtight connection to the FAGE nozzle where OH is measured by LIF (Laser Induced Fluorescence) using the Q1(3) transition (A2Σ+(v=0)-X2Π(v=0)) at 308 nm after gas expansion into a low pressure cell (1.7 mbar). The laser light is generated using a frequency-doubled dye laser (Sirah Laser PrecisionScan PRSC-24-HPR) pumped by the frequency-doubled output of a Nd:YVO4 laser (Spectra Physics Navigator II YHP40-532QW). The laser power used to probe OH was approximately 2 mW.

UL-FAGE is composed of two consecutive low-pressure cells based on the Penn State instrument design (Faloona et al., 2004), one being used to measure OH and the other one to measure HO2. The laser beam is delivered to the OH cell by a set of optical lenses and fibers. Air from the photolysis cell is pumped by a BOC Edwards GX6/100L pump through a 1 mm orifice after gas expansion into the low-pressure cells. The gas expansion makes it possible to freeze chemical reactions and to extend the fluorescence lifetime of OH. The fluorescence signal is measured perpendicular to the excitation beam by a CPM module (Channel Photomultiplier, Perkin Elmer), which detects photons and converts them to TTL pulses. In the OH reactivity mode, only the first cell dedicated to the measurement of OH is used. The pulses generated from the laser-induced fluorescence of OH are counted by an acquisition card (National Instruments) coupled to acquisition software coded in the LabVIEW environment to automate the measurement sequence, to acquire data, and to process the OH reactivity measurements in real time. The data analysis procedure, based on the exponential fit of a defined number of OH decays, is described in Sect. 2.3.3.

The OH decay observed is a sum of losses due to chemical reactions, diffusion out of the probed volume, and wall reactions. Under ambient conditions, the OH concentration will follow Eq. (6): OH=OH0exp⁡-kambient+klossest, kambient and klosses represent first-order losses of OH due to gas-phase and non-gas-phase chemical reactions, respectively. In order to determine the ambient reactivity, two steps are required. First, zero air is introduced in the photolysis cell to quantify the losses (klosses) due to physical processes and wall reactions in the absence of gas-phase chemical reactivity. Then, ambient air is pumped though the photolysis cell to measure the sum of the above-mentioned losses and losses due to the ambient OH reactivity (kambient). The total ambient OH reactivity is the difference between the first order loss rates of OH measured in ambient air (kambient+klosses) and in zero air (klosses).

To measure klosses, approximately 8.3 Lmin-1 of clean air (99.9 %, Praxair 3.0) was provided to the photolysis cell, comprising 6.3 Lmin-1 of dry zero air and 2 Lmin-1 of zero air humidified with high-purity (Milli-Q) water, which provides enough water (approximately 3000 ppmv) for OH generation. The water concentration was measured by a hygrometer (Michell S8000). As mentioned previously, an additional flow of 20 mLmin-1 of dry zero air and ozone is injected into the photolysis cell to maintain a minimum ozone mixing ratio of approximately 60 ppbv in this cell. Care is taken to perform zero measurements at the same working pressure as during ambient measurements. The OH reactivity in zero air was measured once or twice per day and was found to be 6.6±1.2 s-1 (n=16) over the campaign. However, the minimum OH reactivity observed during ambient measurements before the zero subtraction was 4.8  s-1, which suggests that the measured zero may be biased high due to residual impurities in zero air.

Tests performed using synthetic mixtures of VOCs (see Sect. 3.1) started and ended with humidified zero air (from a zero air generator and a water bubbler filled with HPLC-grade water). OH reactivity measurements performed with this humid zero air led to a klosses value of 5.6 s-1, most likely due to the higher quality of the zero air used. For this reason, the klosses value of 5.6 s-1 was used as the pump–probe zero reactivity for the entire intercomparison. It should also be mentioned that during later separated measurements, changing the zero air quality for a higher purity (Air liquid, Alpha1 grade: 99.999 %) led to a lower klosses value of 4.5±0.4 s-1. However, the configuration of the pump–probe instrument (in particular, the laser type and beam shape) was different during the subsequent measurements; thus, the value of klosses measured during the campaign (i.e., 5.6 s-1) has been preferred for the analysis of measurements made during the intercomparison. These results suggest that the measurements reported for the pump–probe instrument may be biased low by approximately 1–2 s-1, due to impurities in zero air, in the water bubbler, and on the Teflon tubing. These observations highlight the difficulty to accurately determine klosses.

As OH is generated by pulse photolysis and is subsequently probed by the FAGE instrument, the OH profile shows a rapid increase followed by a decrease depending on the OH reactivity. Observed decays were multi-exponential, with a first rapid decrease over a few ms, likely due to the fast diffusion of OH within the photolyzed volume before homogenization. This behavior was previously observed for other pump–probe instruments (Sadanaga et al., 2004b; Lou et al., 2010). In order to quantify the OH reactivity, it is therefore necessary to delay the time at which the mono-exponential fit, representative of the chemistry, starts. A discussion on the choice of the mono-exponential fit for derivation of the total OH reactivity is addressed in the following paragraph.

The UL-FAGE instrument is regularly tested using a gas standard, such as carbon monoxide (CO), whose rate constant with OH is well-known (Parker et al., 2011). Monitoring a kinetic parameter allows for the optimization of experimental conditions such as the beam size, the laser power, and the OH decay fitting window. During these tests, CO is introduced in large excess in the cell at different known concentrations. These tests are carried out under NOx-free conditions, and a single exponential decay of the OH signal down to zero is observed after the first few ms. The pseudo first-order rate constants from the measured OH decays are plotted against the CO concentrations; a linear regression of this plot yields the second-order rate constant. The shortest delay between the photolysis pulse and the start of the fit (0.025 s), during which additional physical losses without chemical interest are observed, was determined by comparing the measured and recommended rate constants for the CO + OH reaction (at 298 K and 1 bar), based on a recommended value of 2.31×10-13 cm3molecule-1s-1 (Atkinson et al., 2006).

A delay of 0.025 s leads to a value of 2.41×10-13 cm3molecule-1s-1 for the CO + OH reaction rate constant, determined from OH reactivity values ranging from 10 to 90 s-1 (R2=0.997). This delay has been used during the present study. Varying the delay from 0.025 to 0.07 s leads to a difference of 7 %, which was accounted for in the total uncertainty calculation and considered as the uncertainty due to the choice of the delay. The delay used in this work is higher than the value used in previous studies (0.008 s in Parker et al., 2011) due to differences in the beam shape, the beam expansion, and the alignment. These results stress that a systematic measurement of OH reactivity using a gas standard should be performed before each campaign and ideally during the campaign. For this study, several tests were carried out using complex synthetic mixtures of known OH reactivity (Sect. 3.1.2).

During ambient measurements, the duration of the exponential fit is adjusted depending on the OH reactivity level. kOH is first estimated (kOH,estimated) by fitting the decay on a duration of 0.038 s, starting after the chosen delay. The decay is then fitted a second time on a duration corresponding to 15 times kOH,estimated. A background signal (due to noise) is subtracted considering the signal level after this delay. In practice, the decay is fitted after averaging acquisitions from subsequent photolysis laser pulses (in general 30) to get a signal-to-noise ratio higher than 5.

The performance of the UL-FAGE instrument is similar to that of other pump–probe instruments (Table 2). This instrument can measure OH reactivity values up to 150 s-1. Operating conditions described above lead to a precision (1σ) of 8 % at 20 s-1 (30 s acquisition) and 10 % at 100 s-1 (1 min acquisition). The accuracy is estimated to be 20 % (1σ) from the CO tests. The 3σ detection limit determined from zero air measurements is 3.6 s-1.

The method to generate OH in the pump–probe technique does not lead to HO2 formation. However, HO2 can be formed during the reaction of ambient trace gases with OH in the photolysis cell. It is observed that the OH signal does not decay to zero, but instead a slowly decaying residual signal is observed even at long reaction times (Fig. 4). This residual OH signal is estimated to be 0.03 %, due to the recycling of OH radicals from the reaction of HO2 with NO, and its intensity, generally a few percent of the initial OH concentration, depends on the NO concentration.

The OH decays are analyzed using a mono-exponential fit in this study. In order to estimate the impact of the recycling chemistry mentioned above on the derivation of OH reactivity measurements, detailed modeling has been carried out below for some observed OH decays. To estimate the bias due to the use of a mono-exponential fit, we carried out a comparison of the decay rates resulting from mono- and bi-exponential fits for several OH decays acquired in a range of different NOx concentrations to decay rates derived from model calculations. In Fig. 4 is shown a decay recorded on 23 October at 21:00 LT, a time period characterized by high NO (91 ppbv) and NO2 (38 ppbv) mixing ratios; the residual OH signal due to recycling of peroxy radicals with NO is clearly visible on the logarithmic scale. The bi-exponential least square fit captures the slow decay and closely reproduces the experimental data; taking the fast decay as the OH reactivity leads to a value of 66 s-1 for kOH. A fit to a mono-exponential decay, using only a subset of the data (in this example up to 0.25 ms, obtained such as explained above as 15 times a preliminary decay rate deduced from a linear regression of ln⁡(signal)=f(t) in the time window 0.025–0.063 s) leads to a value for kOH of 59 s-1, i.e. 10 % below the value obtained from the bi-exponential fit.

The same OH decay is shown on the right panel of Fig. 4 together with OH profiles from model calculations. The model used for these calculations is given in the Supplement. OH profiles are shown for two different NO / NO2 mixing ratios: the measured ambient mixing ratios (91/38 ppbv), and mixing ratios (77/51 ppbv) calculated based on the conversion of a fraction of NO into NO2 due to the addition of 60 ppbv of O3 into the photolysis cell and a residence time of 6.6 s. In both cases, the VOC reactivity has been set to best reproduce the measured ambient OH decay. The top trace, simulated using ambient NOx mixing ratios, overpredicts the recycled OH concentration, while experimental data are better reproduced using the NO / NO2 partitioning calculated to be present in the cell. Indeed, the addition of high concentrations of O3 combined with the rather long residence time before intake into the FAGE cell decreases the impact of NO on the retrieved OH decay.

The first fast decay is mainly due to the reaction of OH with primary species. The second, slow decay is mostly governed by the loss of the recycled OH radicals through reaction of RO2 radicals with NO2 and to a lesser extent by the reaction of HO2 with NO2 as well as OH with NO and NO2. Diffusion and wall losses of radicals have very minor contributions. Both modeled traces do not follow a bi-exponential decay, but exhibit a short plateau after the first fast decay before starting the second slower decay. Such a shape reproduces the experimental data very well, as can be seen in particular for the blue trace, but cannot be assimilated to a bi-exponential decay anymore. From the VOC reactivity, used in the model to best reproduce the experimental trace, a kOH of 57 s-1 is obtained, which is closer to the result from the single exponential decay. Using kOH such as that obtained from the bi-exponential decay and introducing a corresponding OH reactivity due to reaction with VOC into the model overpredicts the OH decay (dotted blue line in Fig. 4).

Testing this analysis on different traces characteristic of different NOx levels shows that the mono-exponential fit leads to an underestimation only during extreme conditions (30 % at high NOx mixing ratios: 150 ppbv and high NO / NO2: 2.7), whereas the bi-exponential fit overestimates (by 10–20 %) the reactivity for all the conditions tested by the model (NOx> 45 ppbv, NO / NO2 > 1.2). From these considerations, we have chosen a single exponential decay as the analysis method for this study, even under high NOx concentrations, rather than applying a bi-exponential fit for traces measured under conditions with NO concentrations above a pre-defined threshold, such as proposed by Lou et al. (2010). However, it should be noted that the differences between the mono- and the bi-exponential fits are within the accuracy of the technique. Laboratory experiments are planned in the future to perform a detailed investigation of the impact of this OH recycling on the reactivity measurements.

OH reactivity values measured by the CRM instrument were found to be (39±2) % lower than that calculated for the NMHC mixture (Fig. S8b) and (53±0.4) % lower than calculated for the OVOC mixture (Fig. S8a). These values are greater than the 2σ uncertainties of 24 and 22 % calculated from a quadratic propagation of uncertainties on the rate constants and the VOC concentration measurements for the NMHC and OVOC mixtures, respectively. Uncertainty calculations were performed assuming that errors on rate constants are independent from each other and that errors on measured VOC concentrations are characterized by an independent random error of 5 % and a similar systematic error of 10 % for each VOC that cannot cancel out through the quadratic propagation of uncertainties on individual OH reactivity values. As discussed below, the OH reactivity measured by the pump–probe instrument for the NMHC mixture, however, agreed well with the calculated OH reactivity values. These results suggest that the MD-CRM instrument underestimates the OH reactivity of complex VOC mixtures, while tests performed using ethane, ethene, and propene did not reveal an underestimation (see Sect. 2.2.3).

These mixtures contain compounds, such as aromatics for NMHCs and acetone for OVOCs, that generally have larger absorption cross-sections at 185 and 254 nm than the calibration gases used to characterize the CRM instrument (ethane, ethene, propene) (Tables S1, S2). Following this study, the MD-CRM instrument was modified to reduce the amount of photons leaking into the reactor (Michoud et al., 2015; Zannoni et al., 2015). The new version of this instrument exhibits less than 5 % of photolysis for pyrrole, compared to approximately 15–25 % in this study. Further tests performed on the modified version of the MD-CRM instrument using mixtures of NMHCs and OVOCs showed a good agreement between the measurements and the OH reactivity expected from these mixtures, with differences lower than 9 % (Michoud et al., 2015). It is therefore likely that the CRM underestimation of the VOC reactivity during this intercomparison experiment was due to photolytic processes inside the reactor, and care must be taken to minimize this issue on CRM instruments, to evaluate these processes, and to ensure agreement between measured and calculated reactivity by performing tests with complex VOC mixtures.

It is interesting to note that if the photolysis of VOCs inside the reactor leads to less reactive species, an underestimation of the OH reactivity is expected. However, the photolysis of VOCs will first lead to the formation of two organic peroxy radicals. If these peroxy radicals quickly react with OH (k≈10-10 cm3s-1), as recently measured for CH3O2 (Bossolasco et al., 2014) and C2H5O2 (Faragó et al., 2015), an overestimation of the OH reactivity would be expected. This point will need to be investigated further.

Figure S8 shows a comparison between the measured and calculated OH reactivity values. For NMHCs, the linear regression gives a slope of 0.99±0.01 (1σ) and a R2 value of 0.99, indicating a very good agreement between the measured and calculated OH reactivity for this synthetic mixture. These results confirm that the fitting procedure described above is robust and suggest that NMHC photolysis can be neglected.

Due to technical issues, only two measurements were made with the OVOC mixture. A close inspection of Fig. S8 may suggest that UL-FAGE overestimates the OH reactivity by 30 % for the measurement made at a calculated OH reactivity of approximately 40 s-1. However, only two measurements are shown on this figure, and more experiments are needed to confirm this observation.

An estimation of the VOC fraction that can be photolyzed by the pump laser can be derived from absorption cross sections at 266 nm and the laser photon density. Benzaldehyde is used as an example, as its absorption cross section at 266 nm (2.14×10-18 cm2) (Etzkorn et al., 1999) is among the largest for the VOCs present in the OVOC mixture (Table S2). Taking into account the UL-FAGE operating conditions (20 mJpulse-1), less than 2.6 % of benzaldehyde is photolyzed per laser shot. If the photolysis of benzaldehyde leads to unreactive species, the OH reactivity from this compound will be decreased. Considering the residence time in the cell and the ratio of the photolyzed volume to the total volume, this would lead to a decrease of 7.8 % (0.23 s-1). On the other hand, as mentioned above for the CRM instrument, the photolysis of benzaldehyde could produce two RO2 radicals, which can then quickly react with OH (k≈10-10 cm3s-1). These conditions correspond to the worst-case scenario, since the rate constant of benzaldehyde with OH is slow (1.2×10-11 cm3s-1) (Atkinson and Arey, 2003) compared to the reaction of CH3O2 with OH. The change in the OH reactivity for this compound, considering the OH consumption by RO2 (lifetime shorter than 1 s), would be an increase of approximately 37 %. For 10 ppbv of benzaldehyde, the resultant increase in the OH reactivity would be 1.2 s-1, which is within the measurement uncertainty.

If we consider the highest OH reactivity measurements made during these experiments, the same calculation can be done for all NMHCs/OVOCs. The calculations indicate a decrease of 0.15 s-1 for all NHMCs and 0.32 s-1 for all OVOCs if the species formed are assumed to be unreactive. If the RO2 pathway is considered, this leads to an increase of 0.5 s-1 for all NMHCs and 1.17 s-1 for all OVOCs. These results suggest that photolysis of VOCs inside the reactor has a minor impact on the measurements. However, a more complex chemistry may have to be considered if the overestimation of the reactivity is confirmed with OVOC mixtures.

It is interesting to note that if the regression line for the CRM measurements in Fig. S8b (red squares) is not constrained to zero, there is an intercept of 1.4±1.0 s-1. While the uncertainty on the offset is large, this offset would be consistent with an OH reactivity of 2 s-1 in the zero air. However, it must be noted that any discrepancy in the determination of klosses (discussed in Sect. 2.3.2) should not affect the OH reactivity measured by the pump–probe instrument during these tests. This is because the same zero air that was used to measure klosses was also used to dilute the standard mixtures. Furthermore, the OH reactivity from the zero air was subtracted from all measurements.

For the MD-CRM instrument, C1 was observed to be stable over the course of the 2-week intercomparison with a value of 47.6±2.0 ppbv (Fig. S9), showing a relative standard deviation of approximately 4 %. From measurements of C1 and C0 (as described in Sect. 2.2.2), it was inferred that 25 % of pyrrole is lost when the mercury lamp is turned on. However, as mentioned in Sect. 2.2.2, further testing made after this campaign (Michoud et al., 2015; Zannoni et al., 2015) indicated that this value is an upper limit of the pyrrole fraction that is photolyzed. It is estimated that the pyrrole photolysis was likely lower by 10 % due to residual OH in the reactor during the C1 measurements. However, an underestimation of C1 has opposite effects (though similar in magnitude) on OH reactivity values derived from Eq. (3) and on the correction factor applied for a deviation from first-order kinetics. As a consequence, this error cancels out in the calculation of the corrected OH reactivity measurements, and an increase of C1 by 10 % does not impact the CRM measurements presented in this study.

Field determinations of the correction factor applied to account for a deviation from pseudo first-order kinetics were carried out using ethane at an apparent pyrrole-to-OH ratio of 1.7 and were compared to laboratory results in Fig. 2. A close inspection of this figure shows that the field and laboratory determinations are in good agreement. In order to check the instrument stability, automatic standard additions of ethane were performed every 3 h throughout the campaign (see Sect. 2.2). The concentration of ethane was adjusted to lead to an increase of the OH reactivity by 24.4±5.3 s-1. The average additional OH reactivity from ethane measured for the entire campaign (N=118 additions) was determined to be 26.1±4.0 s-1 (Fig. S10). These results indicate that the CRM instrument ran smoothly during the campaign (i.e., no drift was observed) and also indicate that the OH reactivity from ethane was not significantly impacted by matrix effects.

The impact of the corrections described in Sect. 2.2.3 on the measured OH reactivity values is shown as Fig. S11. The corrections on C2 due to a change of RH between C2 and C3 measurements range from -10 to +15 s-1, with the sign depending on whether humidity was higher or lower during C3 compared to C2. The correction of C3 due to the secondary formation of OH inside the reactor from HO2+NO for NOx mixing ratios lower than 40 ppbv (observed most of the time during the first 13 days of the campaign) has a similar impact on the OH reactivity measurements as the humidity correction, with the correction ranging from 0 to 15 s-1 (7.2±7 s-1 on average). This correction was higher on 22–23 October, when the site experienced unusual elevated NOx mixing ratios of 75–340 ppbv (Figs. 4, S11). These large mixing ratios of NOx led to corrections ranging from 20 to 110 s-1, with higher values during the peak of NOx seen on the morning of 23 October.

The correction applied for a deviation from first order conditions is small and is also well characterized. This correction leads to a 10 % decrease of the measured values. The correction applied for dilution leads to an increase of 20 % of the measurements. Overall, while the amplitude of some of the corrections applied on the CRM measurements are large, a good characterization of each correction makes it possible to correct the measurements without introducing unreasonable uncertainties when NOx mixing ratios are lower than 40 ppbv.

The measurement precision (1σ) derived from measured PTR-MS signals can be computed from (LOD/3) + (5 %×kOH) and is approximately 25, 12, 8, and 7 % at ambient OH reactivity values of 5, 15, 30, and 60 s-1, respectively. Based on a quadratic propagation of errors associated with the PTR-MS calibration, the rate constant for the pyrrole + OH reaction, the correction for a deviation from pseudo first-order conditions, and the dilution, the total measurement uncertainty is computed from (LOD/3) + (18 %×kOH) when NO mixing ratios are lower than 5 ppbv. The measurement uncertainty is 38, 24, 20, and 19 % for ambient OH reactivity values of 5, 15, 30, and 60 s-1, respectively. For NOx mixing ratios ranging from 5 to 40 ppbv that are associated with ambient OH reactivity values higher than 10–20 s-1, the total uncertainty increases by less than a factor of 1.5 and leads to uncertainty values of 20–30 % (see Michoud et al., 2015 for the methodology used to estimate uncertainties). For NOx mixing ratios higher than 40 ppbv, the total uncertainty increases by up to a factor of 3, leading to total uncertainties in the range 20–70 %.

The UL-FAGE pump–probe instrument ran continuously during the campaign. The pump laser was regularly optimized to maintain an energy of 20 mJpulse-1, which was continuously recorded during ambient measurements using a photodiode. However, it is interesting to note that if a drift in the pulse energy leads to a lower S / N ratio, this will not affect the shape of the decay, keeping the measured OH reactivity unchanged. As mentioned previously, the OH reactivity measured in zero air (99.9 % purity) was stable through the entire campaign (N=16, 6.6±1.2 s-1) and indicates that instrumental drifts were not significant. The measurement accuracy was tested using CO mixtures at the end of the campaign and was found to be better than 20 % (1σ). More regular additions of a standard will be performed during future campaigns to track the performances of the UL-FAGE instrument in real-time.

Figure 5 shows the time series of OH reactivity measurements from the CRM and pump–probe instruments together with mixing ratios of total VOCs and NOx. This figure indicates that measured OH reactivity values were in the range 5–100 s-1 with ambient mixing ratios of total VOCs and NOx usually ranging 10–80 and 10–300 ppbv, respectively. Except for the last 3 days of the campaign, where elevated NOx mixing ratios (50–300 ppbv) were observed, these conditions are similar to other urban sites (Dolgorouky et al., 2012; Mao et al., 2010; Shirley et al., 2006).

Diel averages of measurements are shown as Fig. 6. Figure 6a displays average diel profiles of measured reactivity values from both instruments, together with NMHCs, OVOCs and NOx mixing ratios during the weekdays from 8 to 21 October, whereas Fig. 6b shows the same profiles during the weekend. Figure 6c displays the diel averages for 21–23 October, a period that is marked by elevated NOx mixing ratios. Figure 6a–b indicate that two peaks are observed for VOCs and NOx whose timing is consistent with that of peak vehicle traffic (08:30–14:00 and 16:30–21:00), with lower values of OH reactivity and less pronounced peaks on weekends. These peaks are commonly seen in urban environments (e.g., Shirley et al., 2006; Mao et al., 2010; Dolgorouky et al., 2012).

Figures 5 and 6 show that the agreement between the pump–probe and CRM measurements is good from 9 October to 21 October, when ambient NOx mixing ratios are lower than 40 ppbv. Similar trends are observed on the two data sets, and any difference is within the measurement uncertainties. The average diel profiles (Fig. 6a, b) indicate a good agreement for low OH reactivity values when both VOCs and NOx are low and during traffic hours on weekdays, when OH reactivity values up to 20–24 s-1 were measured. In contrast, the agreement is not as good during traffic hours on weekends, which is mainly due to lower CRM measurements recorded on Sunday 21 October 2012 from 19:00 to 24:00. However, this disagreement is within the measurement uncertainties of the two techniques.

At first glance, the relatively good agreement between the two instruments seems to be inconsistent with an underestimation of the total ambient OH reactivity by the CRM instrument due to the photolysis of VOCs inside the sampling reactor as discussed in Sect. 3.2.1. However, a concomitant underestimation of the pump–probe measurements by 2 s-1, due to a zeroing issue as discussed in Sects. 2.3.2 and 3.1.2, may lead to a fortuitous good agreement. For instance, increasing the pump–probe measurement by 2 s-1 at a measured OH reactivity of 10 s-1 would lead to a relative difference of 20 %. Assuming that the CRM instrument underestimates the OH reactivity due to VOCs by 45 % and taking into account that VOCs and NOx exhibit similar contributions to the total OH reactivity (bottom panels of Fig. 6, black trace), an underestimation of approximately 23 % would be expected for the CRM measurements, which is consistent with the underestimation that would be observed if the pump–probe measurements were increased by 2 s-1. Due to the nature of the underestimation for each instrument, the effects on the OH reactivity measurements are not apparent in Figs. 5 and 6. While the underestimation by the pump–probe instrument is essentially constant, the underestimation by the CRM instrument varies with the measured ambient OH reactivity.

For NOx mixing ratios greater than 40 ppbv (22–23 October), the agreement between the CRM and pump–probe measurements is still reasonable (Fig. 5), with values for the CRM instrument that are lower by 30 % on average than those from the pump–probe instrument, mainly due to the large difference observed on the morning of 22 October. It is interesting to note that the correction procedure used for the CRM measurements seems to be accurate although large corrections are applied.

Figure 7 shows scatter plots of the OH reactivity measured by the CRM and the pump–probe instruments for the whole campaign. The time period from 19:00 LT on 21 October to 07:30 LT on 22 October was not included in the scatter plot and is shown using a different color and symbol. A large difference was observed between the two instruments during this time period, but no obvious reasons were found for this disagreement. A linear regression gives a slope of 0.78±0.01 (1σ) and an intercept of 2.16±0.1 (1σ) s-1, consistent with an underestimation of the OH reactivity by the CRM instrument due to VOC photolysis in the sampling reactor and a bias of approximately 2 s-1 in the determination of klosses for the pump–probe instrument. Since the bias in klosses for the pump–probe instrument is independent of the measured ambient OH reactivity, this bias will be reflected in the intercept. Similarly, as the underestimation of OH reactivity by the CRM instrument is relative to the measured OH reactivity, this underestimation will be reflected in the slope. Including the 21–22 October time period in the scatter plot leads to the same slope and an intercept of 2.74 s-1. This linear relationship indicates that for ambient OH reactivity values lower than 10 s-1, the CRM measurements will be higher than the FAGE measurements, while the opposite is expected for higher ambient OH reactivity values.

A color-coding of the OH reactivity measurements shown in Fig. 7 as a function of the ratio of VOC to NOx reactivity also suggests that the CRM measurements are systematically lower than the pump–probe measurements for ambient OH reactivity values lower than 35 s-1 when the OH reactivity is driven by VOCs. It is interesting to note that a color-coding as a function of NO does not indicate a NO-dependent bias and also tends to validate the corrections applied on the CRM measurements.

The results of this intercomparison suggest that some improvements can be made to both instruments. For the pump–probe instrument, a better assessment of the background reactivity is critical to lead to more accurate measurements, especially at low OH reactivity values. High purity water and ultra-high purity air (such as alpha 2 from Air Liquide or, ideally, air produced from liquid O2 and N2) should be used to determine the zero air reactivity.

There are three main areas of improvement for the MD-CRM instrument: (i) a reduction of photolytic processes inside the sampling reactor, (ii) a reduction of the magnitude of the humidity correction, and (iii) a reduction of the magnitude of the HO2 + NO correction. VOC photolysis can be reduced by modifying the geometry of arm 2 (Fig. 1b) to reduce the amount of photons leaking into the reactor. Reducing the humidity correction requires the use of a generator that can supply zero air free of both VOCs and NOx at a humidity level that mimics ambient air. It is interesting to note that while catalytic converters can be used to generate humid zero air free of VOCs at the same relative humidity than ambient air, these converters do not remove NOx species and are not suitable for OH reactivity measurements in urban and suburban areas. The correction to apply for the presence of ambient NOx could be reduced by using a source of OH that does not produce HO2, as the reaction of HO2 with NO is the main source of the NO artifact.

After this study, the MD-CRM instrument was optimized to reduce photolysis inside the reactor by changing the position of the mercury lamp. Pyrrole photolysis was decreased to less than 5 % (Michoud et al., 2015; Zannoni et al., 2015), and tests performed with NMHCs (aromatic compounds) and OVOCs (aldehydes, ketones) indicate that less than 1 % of these compounds are photolyzed (Michoud et al., 2015). In addition, further measurements of OH reactivity from complex synthetic mixtures indicated that the new version of the MD-CRM instrument is capable of measuring the OH reactivity of these mixtures at ±10 %. The catalytic converter used during the intercomparison has been replaced by a home-built air generator fed with dry zero air free of both VOCs and NOx and designed to continuously adjust the RH output to match ambient RH. With this air generator, the humidity correction is decreased below 3–5 s-1 (Michoud et al., 2015).

The reduction of the photon flux inside the reactor also made it possible to reduce NO2 photolysis to negligible levels, which in turn makes the MD-CRM instrument less sensitive to the HO2+ NO artifact. The conversion of NO2 into NO decreased from approximately 50–20 %. Work is ongoing to reduce the last 20 % of conversion, which is mainly due to a heterogeneous conversion on surfaces. Tests are also ongoing to determine whether there is a clean source of OH that could be used on the CRM technique to minimize the formation of HO2, and therefore the correction related to ambient NO.