Peroxy radical partitioning during the AMMA radical intercomparison exercise

. Peroxy radicals were measured onboard two scientiﬁc aircrafts during the AMMA (African Monsoon Multidisciplinary Analysis) campaign in summer 2006. This pa-per reports results from the ﬂight on 16 August 2006 during which measurements of HO 2 by laser induced ﬂuorescence spectroscopy at low pressure (LIF-FAGE) and total peroxy radicals (RO ∗ 2 = HO 2 + 6 RO 2 , R = organic chain) by two similar instruments based on the peroxy radical chemical ampliﬁcation (PeRCA) technique were subject of a blind intercomparison. The German DLR-Falcon and the British FAAM-BAe-146 ﬂew wing tip to wing tip for about 30 min making concurrent measurements on 2 horizontal level runs at 697 and 485 hPa over the same geographical area in Burkina Faso. A full set of supporting measurements comprising photolysis frequencies, and relevant trace gases like CO, NO, NO 2 , NO y , O 3 and a wider range of VOCs were collected simultaneously.


Introduction
Hydroperoxy-(HO 2 ) and alkyl peroxy (RO 2 ; R = organic chain) radicals are involved in many of the oxidation mechanisms taking place in the troposphere. They are short lived species originating from the oxidation of carbon monoxide (CO) and volatile organic compounds (VOCs) (Monks, 2005). Their reactions are closely coupled with those of the most important atmospheric oxidative agent, the hydroxyl radical OH, and the formation mechanisms of chemically or toxicologically important pollutants like ozone (O 3 ), peroxy acetyl nitrate (PAN), aldehydes, acids, etc. Knowledge of the atmospheric radical content and radical distribution provides essential information about the oxidative state of an air mass. In spite of their importance in the chemical processing of the troposphere, only a very limited number of airborne peroxy radical measurements are available (Brune et al., 1998;Jaeglé et al., 2001;Hanke et al., 2002;Cantrell et al., 2003a, b;Martinez et al., 2008). In particular, there are no previous measurements in or over Africa.
Published by Copernicus Publications on behalf of the European Geosciences Union.
Within the European Project AMMA (African Monsoon Multidisciplinary Analysis) the fast photochemical processing during the West African Monsoon (WAM) period was investigated. Overall the AMMA project aims to improve the knowledge about the WAM variability and the impact on the local and global atmospheric composition. One of the main scientific objectives was to investigate the development of Mesoscale Convective Systems (MCS), including their impact on the transport of trace gases. For this, 5 different instrumented aircrafts were deployed within AMMA covering the measurement of trace gases over a broad vertical and horizontal area . The importance of convection in determining the abundance of radical precursors in the Upper Troposphere (UT) has recently been confirmed experimentally within AMMA by Bechara et al. (2009), who measured enhancement of isoprene and other VOCs up to 10 km altitude within MCS episodes. Some of the products of isoprene oxidation have lifetimes of several hours, and it is possible that such compounds, as well as brominated and iodinated hydrocarbons, are locally transported to the UT by convective activity. Peroxy radical observations made within AMMA (Andrés-Hernández et al., 2009) are indicative of enhanced in situ production of peroxy radicals in the upper troposphere during convective episodes, linking the boundary layer, free troposphere and UT/LS chemistry.
The general interpretation of the impact of radical chemistry relies consequently on the data quality control designed to prevent any misinterpretation of results derived from offsets or other kind of instrumental artefacts. The intercalibration of similar instruments is a crucial part of the data analysis and interpretation, and the results obtained can be used to provide an estimate of the uncertainty associated with the airborne measurement of radicals. As reported by Reeves et al. (2010) wing tip to wing tip comparison flights along straight and level runs at various altitudes were performed during AMMA. The present work focuses on the comparison and evaluation of peroxy radical measurements.

Experimental description
Three instruments for the measurement of peroxy radicals participated in the intercomparison exercise which took place on the 16 August 2006. Figure 1 depicts the corresponding flight track over the MSG (Meteosat Second Generation) satellite image at 15:00 h with IR brightness temperature. These images were usually used to locate the MCS during AMMA. The German DLR-Falcon (hereafter referred to as D-Falcon) and the British FAAM-BAe-146 (hereafter referred to as BAe-146) flew wing tip to wing tip for approximately 30 min, first in a straight and level westward run at 697 hPa followed by a straight and level eastward run at 485 hPa over the same geographical area in Burkina Faso. More details of the intercomparison including the results for measurements of CO, O 3 , CO 2 , HCHO and NO y are given  in Reeves et al. (2010). Results of the initial CO comparison led to a comparison of standards which showed that the data for the D-Falcon and BAe-146 should be corrected by a factor of 0.997 and 1.069, respectively. Table 1 summarises the instrumentation in both aircrafts used for the present data analysis. A more detailed description of individual instruments is included in the following sections

Peroxy Radical Chemical Amplification
The Peroxy Radical Chemical Amplification technique (PeRCA) utilises the amplified conversion of the peroxy radicals entering the reactor into NO 2 in a chain reaction involving NO and CO, namely: By switching the addition of the reagent CO with an identical flow of N 2 between the front and back of the reactor the chain chemistry is modulated between chain amplification mode (i.e., NO 2 from the conversion of peroxy radicals plus other background trace gases producing NO 2 in their reaction with NO like O 3 , plus NO 2 itself) and background measurement mode (i.e., NO 2 comprising that in outside air plus Atmos. Chem. Phys., 10, 10621-10638, 2010 www.atmos-chem-phys.net/10/10621/2010/ other background trace gases producing NO 2 in their reaction with NO). NO is added continuously at the front of the reactor, such that in amplification mode NO and CO are added together and N 2 is added at the back, while in background mode NO and N 2 are added together with CO added at the back. This methodology maintains the overall properties of the reagent mixture in the sample airflow downstream of the reactor and reduces pressure pulsing on switching measurement modes.
The chemiluminescence of the reaction of NO 2 with luminol (3-aminophthalhydrazide: C 8 H 7 N 3 O 2 ) is used to measure the NO 2 . The amount of peroxy radicals is calculated from the difference between the signals in both modes, NO 2 , and the knowledge of the amplification factor or chain length (CL) and radical loss at the inlet. PeRCA measures the total sum of RO, OH, HO 2 and RO 2 radicals. As [RO 2 ] [RO] and [HO 2 ] [OH] in the troposphere, the measured signal is a good approximation of the total sum of peroxy radicals (RO * 2 = HO 2 + RO 2 ).
The CL depends on humidity (Mihele and Hastie, 1998;Mihele et al., 1999;Salisbury et al., 2002;Reichert et al., 2003), residence time in and shape and material of the reactor, and on the ratio of the gases added for the chemical conversion, i.e., [NO]/ [CO]. Instruments based on PeRCA have been developed for the airborne measurement of peroxy radicals (Green et al., 2003). In order to increase sensitivity and accuracy, dual systems comprising two identical reactors, which concurrently measure in the same or alternate modes, are coupled with one or two detectors (Cantrell et al., 1996;Green et al., 2006). Interferences related to short term variations of concentrations during the airborne sampling cycle are thereby minimized.

DUALER (DUal channel Airborne peroxy radicaL chemical amplifiER)
As indicated in Table 1, the Institute of Environmental Physics of the University of Bremen (IUP-UB) participated in the wet monsoon measurement campaign of AMMA in August 2006 with the measurement of the total sum of peroxy radicals, RO * 2 , onboard of the German D-Falcon using a DUALER.
The airborne DUALER deployed on the D-Falcon has been developed by IUP-UB from the first calibrated PeRCA (Hastie et al., 1990) which has been gradually improved and implemented in several campaigns and comparison exercises for ground based measurements (Zenker et al., 1998;Andrés Hernández et al., 2001;Burkert et al., 2001aBurkert et al., , b, 2003. This instrument bases on the concept described in Sect. 2.2. for aircraft measurements. The IUP-UB DUALER utilises a common inlet, and comprises a double reactor (21 mm ID, 310 mm L stainless steel cylinders, with a 10 mm inlet orifice) sampling from a common pre-reactor nozzle and connected to two identical luminol detectors. Three way valves in the inlet system are switched every 60 s and alternate the modes of the reactors. The DUALER is kept at a below ambient, constant fixed pressure during the measurement. The CL decreases with the pressure and the sensitivity of the instrument reaches a maximum for an operating pressure around 500 mbar. During the intercomparison exercise the DUALER was kept at 300 mbar. The characterisation, calibration procedures and performance during the measurement campaign are described in detail elsewhere (Kartal et al., 2010).
The known dependency of the CL on the relative humidity (RH) of the air sampled (Mihele and Hastie, 1998;Mihele et al., 1999;Reichert et al., 2003) has a negligible effect under the AMMA measurement conditions. The DUALER inlet is not heated during the AMMA flights but its inner part is connected to the cabin of the aircraft. As a consequence, the temperature of the reactors remains always higher than that of the outside air. Therefore, the RH, being RH= P H 2 O partial /P H 2 O saturation , decreases in the reac-tors respect to the ambient as T ambient <T reactor causes the increase in P H 2 O saturation , and P ambient >P reactor causes the decrease in the P H 2 O partial in the reactors (Kartal, 2009). The in flight detection limit and accuracy of the instrument depend on the stability of the luminol flow, the variability of the air mass composition sampled and the operating pressure. For most of the AMMA airborne measurements the relative accuracy was between 25 and 45% and the RO * 2 detection limit varied between 2 and 5 pptv for a 20 s time resolution (Kartal et al., 2009). During the intercomparison exercise the decrease in the RH relative to the ambient was dominated by the decrease in the P H 2 O partial as the DUALER operated at 300 mbar. The RH in the reactors remained below 15%. At 300 mbar this effect in the CL is expected to be within the error of the CL determination.

PeRCA 4
The University of Leicester airborne PeRCA instrument (designated PeRCA 4) is a dual channel (two inlet, two detector) PeRCA and was operated on board the BAe-146 during AMMA. The PeRCA 4 instrument was originally designed and built jointly between the University of Leicester and the University of East Anglia. The application of the PeRCA 4 dual channel system to ground based measurements was first described by Green et al. (2006). The aircraft instrument, its characterisation, calibration procedures and performance during the measurement campaign are detailed in Brookes (2009).
PeRCA 4 uses two equivalent channels each comprising a low-volume pyrex inlet coupled to a modified LMA-3 Scintrex NO 2 detector channel. A sample flow of 2 slpm is drawn through each inlet inclusive of ambient air and reagent gases. In each inlet the chain chemistry described in Sect. 2.1 is modulated between chain amplification mode and background measurement mode by switching the addition of a flow of the reagent CO (100 sccm, 5%/vol. equivalent in the sample flow) with an identical flow of N 2 between the front and back of the inlet (typically every 60 s)while a flow of the reagent NO (10 sccm, 3 ppmv equivalent in the sample flow) is added continuously at the front of the inlet. Reagents and N 2 are supplied under pressure from gas cylinders held within the instrument rack inside the aircraft cabin with flow control maintained using mass flow controllers (Tylan FC260, now manufactured and distributed by Celerity Inc.). Total sample flow is maintained at 2 slpm (1.79 slpm ambient air, 0.21 slpm reagent gases) by a combination of mass flow meters (MFM, Tylan FM360) in series with stepper motor controlled needle valves; active control of the stepper motors via computer software provides sample flows to within The channels are run out of phase as shown in Fig. 2, to provide simultaneous measurements of chain amplified and background NO 2 , and the radical chain amplified NO 2 signal is determined by subtraction of the background measurement. It was not possible to perform in flight CL calibrations on board the BAe-146, and consequently calibrations were performed at ground level in the laboratory following the AMMA campaign. The calibration procedure for the PeRCA 4 inlet chain lengths is based on the method of Clemitshaw et al. (1997), whereby known concentrations of methyl peroxy are generated through UV photolysis of methyl iodide. The calibration procedure as yet does not take into account the change in CL with temperature and pressure although it would be preferable to perform CL calibrations at different altitudes, to account for the temperature and pressure dependent rate constants of the chain chemistry reactions.
The PeRCA 4 inlets ( Fig. 2) sample the airflow at ambient air pressure perpendicular to the aircraft direction of flight. The inlets can be heated to reduce the impact of humidity on the chain length, however this system fails in flight on the BAe-146. The result is that for the PeRCA 4 instrument the air temperature, pressure and humidity are expected to remain at ambient levels on sampling and impact on the chain length. A systematic reduction in the chain length with altitude has been modelled (Brookes, 2009) using the FAC-SIMILE package (v4.40.101, MCPA Software Ltd.) using an inlet chemistry reaction mechanism based on that of Hastie et al. 1991) and rate coefficients taken from the recommendations of Sander et al. (2006) and Atkinson et al. (2004Atkinson et al. ( , 2006. Pressure and temperature were constrained in the model using the pressure and temperature altitude profiles measured from the BAe-146 during AMMA, further details of the model setup can be found in Brookes (2009). At the altitude of the intercomparison level run where the PeRCA 4 operated (700 mbar) a reduction in the CL of the order 20% is modelled. The reduction of the PeRCA 4 chain length with altitude will contribute to systematic bias between the PeRCA 4 and DUALER measurements, where the DUALER reactor is pressure controlled and the PeRCA 4 chain length has not been corrected. The modelled variation in the CL at altitude (CL alt ) relative to the CL at ground level (CL 0 ) is shown as a function of pressure in Fig. 3.
To correct for the humidity interference to the chain chemistry a correction function (dependent on relative humidity) was derived and applied to the AMMA PeRCA 4 [RO * 2 ] dataset as demonstrated in Fig. 4. The model to measurement ratio as a function of temperature shows the humidity corrected data consistently over estimates the model with no real trend. A slight increasing trend with temperature is observed in the non humidity corrected model to measurement ratio. This has no relationship to humidity and in fact can be explained the changing composition of the radical population with altitude (Brookes, 2009). Model to measurement agreement improves at higher altitude (lower temperature) where the radical population is primarily composed of HO 2 and CH 3 O 2 . At lower altitudes (higher temperature) a more complex population of organic peroxy radicals influenced by the surface emissions of isoprene is modelled and the increased model to measurement ratio reflects a reduced sensitivity of the measurement to these components of the radical population. On application of the correction to the AMMA dataset as whole the concentrations typically scaled on average by a factor of 4.4 resulting in up to several hundreds of pptv [RO * 2 ] (Brookes, 2009); this scaling is considered unrealistic since concentrations this high have not been reported in the tropical troposphere. By comparison of the uncorrected PeRCA 4 [RO * 2 ] dataset to output from the DSMACC model (Emmerson and Evans, 2009;Stone et al., 2010), the uncorrected data was assessed for any systematic bias with humidity (Brookes, 2009). The model to measurement ratio exhibits no trend with relative humidity. To apply the humidity correction to this dataset may therefore be inappropriate and the intercomparison was made using the uncorrected data.
The expected humidity bias in the measurement response was not observed in the PeRCA 4 AMMA dataset despite the consistently high humidity encountered during the monsoon season (Brookes, 2009), and thus a potential explanation is sought. Owing to their low thermal mass the PeRCA 4 inlets are expected to be in thermal equilibrium with the aircraft cabin and ambient airflow. A temperature gradient is also expected between the inlet tip in contact with the ambient airflow, and the inside in contact with the aircraft cabin. The most significant interaction between the inlet and the ambient airflow will occur on sampling and it is hypothesized that at this cold point water vapour may condense, thus reducing the relative humidity sampled. This hypothesis also provides an explanation of the observed loss of sample flow through the mechanism of inlet tip icing; the sample flow was observed to drop dramatically at altitudes above the freezing point in highly saturated conditions during AMMA (Brookes, 2009). Although the condensation of water vapour prior to the reaction zone may reduce the impact of humidity on the chain length, significant sampling losses of HO 2 are expected (Brookes, 2009) as reported by Green et al. (2003). Systematic differences in the RO * 2 measurements may then be attributed to the relative sensitivity of the respective PeRCA instruments to inorganic and organic peroxy radicals.
The PeRCA 4 inlet chain lengths were calibrated following the AMMA campaign in reference to known concentrations of the methyl peroxy radical, for details see e.g. Green et al. (2003). The accuracy of the derived RO * 2 concentration will depend on the relative proportion of the radical population made up by the methyl peroxy radical and radicals with a similar effective chain length. The calibration procedure for the NO 2 sensitivity of the LMA-3 Scintrex detectors is detailed in Brookes (2009). Ordinarily the LMA-3 detectors are calibrated by mixing the output from a VICI Metronics NO 2 wafer permeation device with varying flows of zero air, but owing to technical problems with the calibration system on board the BAe-146, it was not possible to perform reliable in-flight calibrations. An alternative method was developed using the background signal on the LMA-3 detectors along with the in-situ ozone measurements made on board the BAe-146 with a commercial Thermo Environmental Instruments Inc. 49C ozone analyser. The relative levels of ozone and NO 2 encountered during AMMA were such that the background signals measured on the LMA-3 detectors were almost entirely due to the oxidation of reagent NO by ozone with a negligible contribution from ambient NO 2 . The absolute NO 2 background concentration generated from this reaction was derived using the temperature dependence of the reaction rate constant and the variation of the residence time with temperature and pressure. The derived background NO 2 concentration within the sample flow was used to weight the detector response. Thus the NO 2 sensitivity of the detectors was calculated along the flight track for every point during which ozone measurements were available.
The estimated accuracy of the PeRCA 4 measurements is of the order 40%, not accounting for uncertainty introduced by variation in the CL by altitude and systematic differences introduced by the weighting of the response by the CL with respect to the methyl peroxy radical. The average detection limit of the PeRCA 4 dual channel measurement over the intercomparison run at 697 hPa is 10.4 pptv for a 60 s averaging interval.

Fluorescence Assay by Gas Expansion (FAGE)
Fluorescence Assay by Gas Expansion (FAGE) is a low pressure laser-induced fluorescence technique used for the detection of OH and HO 2 . The instrument and associated calibrations are described in detail in Commane et al. (2010) and only a brief overview will be presented here. During Atmos. Chem. Phys., 10, 10621-10638, 2010 www.atmos-chem-phys.net/10/10621/2010/ the intercomparison flight described here, only HO 2 was observed. OH was not observed due to technical problems during the period of the campaign. In this instrument HO 2 is titrated to OH by the addition of pure NO (99.999%). Air was sampled from a side-mounted inlet through a 0.75 mm pinhole into the fluorescence cell. A Nd:YAG pumped Ti:Sapphire laser is used to generate light at λ≈308 nm in order to excite the Q1(2) transition of the OH molecule. The exact excitation wavelength is determined using a reference cell. The subsequent fluorescence of the OH molecule is also detected at λ≈308 nm. Maintaining the fluorescence cell at low pressure (1.5-2.5 Torr) allows for temporal gating of the fluorescence detection. The sensitivity of the instrument to HO 2 was determined by calibration before and after the campaign. In these calibrations, a known concentration of HO 2 is produced by the photolysis of water vapour (at λ=184.9 nm) in a turbulent flow of air. The HO 2 data was collected at 1 Hz and averaged to 60 s for comparison here. The mean limit of detection of the instrument for the data discussed here is 0.31 pptv (60 s integration time).The total uncertainty of the measurement is 26% at 2σ .

Further instrumentation
A complete description of the aircraft instrumentation during AMMA can be found elsewhere . For the data analysis of the present work, O 3 , CO 2 , CO, HCHO and reactive organic compounds measured on board the D-Falcon by using respectively UV absorption , IR absorption , UV fluorescence (Gerbig et al., 1996), a fluorometric technique (Hantzsch reaction), and two O 3 /NO-chemiluminescence detectors and a gold converter for reduction of higher oxidized NO y compouds to NO Ziereis et al., 2000) were used. The NO x was also measured by the UEA NO xy instrument on the BAe-146 that thermally converts NO y to NO 2 , using a gold convertor, photolytically converts NO 2 to NO and measures NO by chemiluminescence Stewart et al., 2008). The instrument has detection limits of 3, 15 and 10 pptv and precisions of 10, 30 and 21% for NO, NO 2 and NO y , respectively.
Information regarding VOCs comes from the 15 min whole air samples (WAS) on board the BAe-146 which were analysed on the ground by dual-channel gas chromatography and flame ionization detection (Hopkins et al., 2003). The VOC detection limits and precisions varied from 1-10 pptv and 1-3%, respectively, depending on compound. Isoprene, acetone and the sum of methyl vinyl ketone (MVK) and methacrolein (MACR) were also measured by proton mass transfer mass spectrometry (PTR-MS) at a higher time resolution and frequency (Table 1), but with detection limits of 20-80 pptv and 10% precision. PAN was measured by gas chromatography (30 pptv detection limit and 3% precision).

Supporting calculations or modelling
Model calculations have been made using the Dynamically Simple Model of Atmospheric Chemical Complexity (DSMACC) (Emmerson and Evans, 2009;Stone et al., 2010). DSMACC is a zero-dimensional box model using the Kinetic Pre-Processor (KPP) (Sandu et al., 2006), constrained to observations of long-lived species with a chemistry scheme generated by the Master Chemical Mechanism (MCM) (Saunders et al., 2003).
A full description of the model calculations undertaken for the AMMA campaign can be found in Stone et al. (2010). To be modelled, time points needed to have observations of HO 2 or RO 2 together with the physical state (latitude, longitude, pressure, temperature and water vapour) and observations of CO, O 3 , NO, j (O 1 D) and j (NO 2 ).
Observations of ethane, propane, iso-butane, n-butane, ethene, propene, acetylene, methanol, ethanol, acetone, formaldehyde, PAN, methyl ethyl ketone (MEK), isoprene, MVK and MACR were also included where available and kept constant at the observed value. All observations used as input to the model were made on the BAe-146. There was virtually no difference in the model output when including benzene and toluene measured on the BAe-146 in test runs. The results indicate that for a set of reactive species the model is relatively insensitive and the model covers the dominant radical production and loss processes.
Concentrations of methane and hydrogen were kept constant at values of 1770 ppmv and 550 ppmv, respectively.
Where observations of alkanes and alkenes were not available, concentrations were estimated as a linear function of the CO concentration .
For each time point, those species which have not been observed are set initially to zero. Those species which have been observed are set to the observed value and kept constant. However, concentrations of oxides are nitrogen are treated differently to other species in the model. For each time point during a day the concentration of NO y (defined as NO + NO 2 + NO 3 + 2 N 2 O 5 + HONO + HNO 4 ) is corrected so that the total NO y concentration is kept constant, but the concentration of the NO y species varies owing to changes in photolysis rates etc. At the end of each day the calculated concentration of one of the NO y species (typically either NO or NO 2 ) is compared to the equivalent measured concentration and the concentration of all NO y species is fractionally increased or decreased so that the measured and the modelled concentrations of the observed species match.
Photolysis rates in the model are calculated from the Tropospheric Ultraviolet-Visible Model (TUV; http://cprm.acd. ucar.edu/Models/TUV/), assuming a constant ozone column of 260 Dobson units and an albedo of 0.1. Cloud correction factors for all reactions other than O 3 →O( 1 D)+O 2 are calculated from the ratio of observed to simulated j (NO 2 ), with the cloud correction factor for O 3 →O( 1 D)+O 2 calculated from the ratio of observed to simulated j (O 1 D).
A removal reaction was included for all species which can be considered as a continuous deposition process (such as dry deposition or continuous wet deposition). The deposition lifetime was initially set at 24 h and the sensitivity investigated.
For each observation of HO 2 or RO 2 the objective is to calculate an expected value from the other parameters measured on board the aircraft and the chemical scheme used. All aircraft parameters have been merged onto a 60 s timebase for expediency. The model is integrated forward in time with diurnally varying photolysis rates until a steady state is reached when the fractional difference in mean species concentration between a day and the previous day is less than 0.01. Full details can be found in Stone et al. (2010). Figure 5 shows the peroxy radicals measured by the DUALER, PeRCA 4 (RO * 2 ) and FAGE (HO 2 ) at the two pressure levels of the intercomparison exercise. As explained in Sect. 2.2.2, a relative humidity correction was applied to the PeRCA 4 data set [RO 2 *-PeRCA 4] but this lead to spuriously high concentrations. Figure 6 illustrates the loss in PeRCA 4 sample flow during the intercomparison flight. With the aircraft ascending to higher altitude the temperature falls below freezing and simultaneously the water content changes from a mixed phase to condensed water (ice) only, as indicated by the Nevzorov probe data from the BAe-146.

Discussion of results
Atmos. Chem. Phys., 10, 10621-10638, 2010 www.atmos-chem-phys.net/10/10621/2010/   The simultaneous drop in sample flow with a phase change in water is seen to indicate condensation followed by freezing occurs, which then blocks the sample flow to the inlet. The combination of the ambient temperature dropping below freezing and elevated ice content coincident to a drop in sample flow is common in all AMMA flights where the PeRCA 4 instrument was operational on board the BAe-146. There has as yet been no experimental verification to quantify the level of humidity within the inlet in flight.
[RO 2 *-PeRCA 4] are only available for the first pressure level measured (697 hPa) owing to a loss of sample airflow at the higher altitude. As can be seen in Fig. 5 For this leg [HO 2 ] varies within and 6 and 18 pptv and the temporal variation shows two abrupt steps at 15:35 and 15:42 h. Both of these steps follow periods where the laser wavelength is rescanned to the maximum of the OH rotational line. This rescanning process takes ca. 2 min and no ambient data are collect during this procedure. Laser diagnostic measurements of power and wavelength stability during the three HO 2 measurement periods in this leg show there is no evidence to suggest that the instrument sensitivity has undergone a change and it is concluded that the two steps in the HO 2 concentration are real variations. It can be demonstrated that the observed [HO 2 ] is directly and linearly proportional to the excitation laser power which is monitored at 1 Hz during ambient measurements and used to normalize the observed signal. This normalization procedure has been applied to the data in question. The relationship between the observed [HO 2 ] and excitation wavelength is non-linear and is an integration of the laser wavelength and line width and the OH rotational line width. A confirmation that the laser wavelength was centered on the OH rotational line throughout this period before and after the rescanning process is obtained by monitoring the signal produced in the OH reference cell in which a high and very stable [OH] (produced by the thermolysis of water vapour) is probed by a measured laser power. The signal obtained from the reference cell during the period where the steps in ambient [HO 2 ] were observed was stable. The additional factor which affects the observed [HO 2 ] is the flow rate of NO into the detection cell, and again the relationship is non-linear, but the flow of NO In principle, the atmospheric radical variability is driven by radiation and/or other trace gas variability. Therefore, the radical temporal evolution observed during the flight was compared with the variability of the photolysis frequencies and radical precursors measured simultaneously.

Photolysis rates
Measurements of j (NO 2 ) and j (O 1 D) are presented in Fig. 7. There is no BAe-146 data for the second intercomparison level as the field of view of the downward facing radiometers were obscured owing to the deployment of a protective shutter.
j (O 1 D) was not measured on board the D-Falcon. At 697 hPa, a systematical difference between j (NO 2 ) radiometers on both aircrafts was identified (see Fig. 7). While the downward facing j (NO 2 ) radiometers show very good agreement (within 6%), the D-Falcon upward j (NO 2 ) radiometer is about 25-35% systematically higher than the BAe-146 upward facing sensor at both pressure levels. According to Volz-Thomas et al. (1996), temperature variations between 25 and −40 • C can only account for 2-3% changing in the instrument sensitivity. The calibrations of the upper and lower j (NO 2 ) radiometers in the BAe-146 before and after AMMA are consistent with one another having both been referenced to the TUV radiation model. Calibrations of the D-Falcon sensors carried out before and after the AMMA campaign show little variance (6%). The downward facing radiometers are mainly exposed to indirect diffuse radiation while the upward facing radiometers get both, direct and diffuse radiation. As a consequence, upward looking radiometers have an additional uncertainty due to the azimuth and zenith dependence of the radiometer sensitivity. Previous investigation has shown that the total combined accuracy of the D-Falcon upward looking radiometer is 14% and of the downward looking radiometer is 3%. If both radiometers are subject to similar errors, it is reasonable that the upward facing j (NO 2 ) measurements, with lower accuracy, show a larger difference within the combined total errors, than the downward facing radiometers. Differences in the surface reflectivity around the sensors in each aircraft may also affect the result.
The temporal variations observed in j (NO 2 ) and j (O 1 D) agree qualitatively with the decrease in radical production measured between 15:05 and 15:10 h at 697 hPa but do not correlate at all with the radical variability observed at 485 hPa.

Radical related trace gases
Concerning other related trace gases, NO and NO x remain around the detection limit. Both NO y (D-Falcon) and PAN (BAe-146) are significant higher at 485 hPa (Stewart et al., 2008)  Acetone measured by PTR-MS shows minor short term variations in both pressure levels, and similarly to the HCHO measured on board the D-Falcon decreases slightly at 495 hPa on average from 1 ppbv to 0.8 ppbv. The corresponding WAS results indicate more marked peak concentrations at 15:30 in acetone and also in acetaldehyde which likely result from the integration related to the lower time resolution of the WAS samples.
Owing to the strong altitude dependence on the sampling and zeros reported by Murphy et al. (2010) during the AMMA measurement campaign, no PTR CH 3 OH data are available for the intercomparison period. The GC analysis of the WAS samples at lower time resolution shows a significant CH 3 OH maximum (1.6 ppbv) at the beginning of the first pressure level (around 15:00 h) and an increase from 0.3 to 0.5 ppbv between 15:30 and 15:35 h at 485 hPa. Similarly, MEK increases from 0.05 to 0.7 ppbv at 15:35 h. MEK is an indicator of industrial and gasoline engine emissions, and its major sink is the reaction with OH producing acetaldehyde. Apart from biogenic emissions, CH 3 OH is produced by reaction of the methylperoxy radical with itself and other organic peroxy radicals (Madronich and Calvert, 1990;Tyndall et al., 2001): The HCHO measured on board the D-Falcon increases concurrent to CH 3 OH from 350 to 385 pptv. This can be considered as an upper limit of the CH 3 OH which can be attributed to Reaction (R1b). Other minor sources of methanol are biomass burning and biofuels and fossil fuel combustion and industrial activities, as in the case of MEK. Acetonitrile and benzene do not provide any indication of recent biomass burning influence in the air mass. Summarising, the variability of the trace gases measured at the second pressure level of the intercomparison exercise indicate changes in the composition of the air mass sampled which can reasonably be correlated to the variability of the local peroxy radicals measured.   Figure 9 shows the ratio ([RO * 2 -DUALER]/[HO 2 ]) for both measured levels. At 697 hPa, both the variability of the measured mixing ratios and the number of HO 2 measurements available is smaller and the ratio seems to remain quite stable around 2.5. This ratio is more scattered at 485 hPa, remaining between 1 and 3 as result of the variability of [RO * 2 ] discussed above. There are only a few [RO * 2 ]/[HO 2 ] values reported for the upper troposphere (1.86: Cantrell et al, 2003a;1.25: Cantrell et al., 2003c). It is important to emphasise that [RO * 2 ]/[HO 2 ] is not expected to be constant but to be directly related to the NO and VOC content and history of the air mass sampled.

Radical partitioning
A similar comparison between [HO 2 ] and [RO * 2 ] measured by LIF and Chemical Ionisation Mass Spectrometry (CIMS) respectively was undertaken within the TRACE-P measurement campaign (Eisele et al., 2003;Cantrell et al., 2003a, b ] ratio is not necessarily constant during the three sets of intercomparison flights, but can reasonably be fitted by a slope of 2.46. The CL of the DUALER measured in the laboratory is an effective CL (eCL) coming from the chain reaction of the radicals not being lost in the pre-reactor nozzle. As the peroxy radicals are reacting rapidly with the walls, the losses at this nozzle before the addition of CO and NO are expected to be important. According to the laboratory characterisation of the response of the DUALER for different peroxy radicals (Kartal et al., 2010), higher eCL is to be expected for CH 3 O 2 than for HO 2 as a consequence of lower wall losses of CH 3 O 2 in the pre-reactor nozzle. Using a common eCL for the determination of [RO * 2 ] in ambient samples as during AMMA can cause an overestimation depending on the [HO 2 ]/[CH 3 O 2 ] ratio. The [RO 2 ]/[HO 2 ] ratios vary mostly between 1:1 and 2:1 during the intercomparison exercise and CH 3 O 2 is expected to constitute 88% of the total organic peroxy radicals (see Sect. 4.4). Under these conditions the experimental [RO * 2 -DUALER] reported in this work might be 8 to 16% overestimated.
Radical partitioning over the whole AMMA campaign derived from the simultaneous measurement of [RO * 2 ] and [HO 2 ] on board the BAe-146 is discussed more in detail in Brookes (2009).

Comparisons between modelled [HO 2 ] and [RO *
2 ] with observations are shown in Figs. 10-12. As mentioned above, HO 2 was only partially measured at 697 hPa and RO * 2 -PeRCA 4 is completely unavailable for the second intercomparison period. This reduces drastically the statistics and representativeness of the results obtained within the intercomparison exercise.
At a first glance the model results in Fig. 10 seem to generally overestimate [HO 2 ] during the intercomparison exercise, which results in some overprediction of total modelled [RO * 2 ]. It is possible that flying adjacent to the D-Falcon may have had some effect on the flow conditions within the FAGE inlet, resulting in greater loss of HO 2 in the inlet and reduced HO 2 detection compared to calibration conditions, although such effects have not been reported in previous aircraft HO x intercomparison exercises (Eisele et al., 2003). This might partly explain the stepwise temporal evolution of [HO 2 ] observed at 485 hPa.
It is important noting that the use of the model in particular for the AMMA intercomparison period is subject to important restrictions derived from following issues: Atmos. Chem. Phys., 10, 10621-10638, 2010 www.atmos-chem-phys.net/10/10621/2010/ Model results (triangles) are constrained to j(NO 2 ) and j(O 1 D) measured on board the BAe-146.  Model results (triangles) are constrained to j(NO 2 ) measured on board the D-Falcon scaled to the j(NO 2 )-BAe values measured at 697 hPa. (a) Photolysis rates: As described in Sect. 3, the DSMACC model requires observations of j (NO 2 ) and j (O 1 D), which were not available for the second altitude level. The existing discrepancies between j (NO 2 ) measured on board the BAE-146 and D-Falcon prevent the direct use of the j (NO 2 )-D-Falcon data for the 485 hPa pressure level without introducing unknown uncertainties. Model runs using j (NO 2 )-D-Falcon show higher radical concentrations than model runs constrained to photolysis rates measured on the BAe-146 as expected since the j (NO 2 )-D-Falcon measurement is systematically 25% higher than the j (NO 2 )-BAe-146 measurement (see Fig. 7). Scaling of the j (NO 2 )-D-Falcon to the values observed on the BAe-146 gives similar results to model results constrained to the BAe-146 photolysis results, and provides some additional data points during the second intercomparison period (Fig. 11). The agreement between [HO 2 ] modelled and measured improves and the modelled [RO * 2 ] for the second period show good agreement on the average with the observed [RO * 2 -DUALER] though the short term variability of the observations is not completely picked up (Fig. 12).  Concerning j (O 1 D), the BAe-146 measurements may have been subject to a systematic bias derived from the non-linearity of j (O 1 D) calibration factors for total O 3 column, temperature and solar zenith angle, as described in Stone et al. (2010) and Brookes (2009). This results in a potential requirement for reduction in j (O 1 D) by a factor of 0.7. The bias appears more significant at altitudes above 4 km where the impact of cloud and aerosol is much reduced. A similar correction to the j (O 1 D) in the model did result in a lowering of the modelled HO 2 , and although the reduction in j (O 1 D) rectifies much of the overestimation for HO 2 in the model, it was not sufficient to explain the model discrepancy during the intercomparison period.
(b) Input of trace gas observations: There are a limited number of VOC observations at high temporal resolution during the intercomparison period. As the input for the model is updated every 60 s, the use of WAS samples with 15 min resolution can introduce unrealistic stepwise significant variations in the mixing ratio of trace gases. In addition, as reported on Reeves et al. (2010) the HCHO measured on board the BAe-146 showed unrealistic oscillations during the intercomparison exercise. Figure 13 compares the HCHO output of the model with the HCHO measured on board the D-Falcon. Furthermore, as the NO x mixing ratios remained below the detection limit of the instruments deployed, the comparison of the model with a reliable experimental reference as described in Sect. 3 is not possible.
The importance of some of these issues in the application of box models to observationally based predictions of HO x radicals has been recently evaluated by Olson et al. (2004Olson et al. ( , 2006. The present work intends to focus on the analysis of the uncertainties of experimental data rather than on the comparison with model results. Given the limitations described above and the inherent challenges of the airborne radical measurement and modelling, the overestimate of HO 2 during the intercomparison period does not appear to be an indication of an inability in the model to replicate observations of HO 2 . In fact there is no significant overestimation of measured [HO 2 ] for the modelling of the AMMA campaign as a whole, as discussed in Stone et al. (2010) and summarised in Fig. 14. The major uncertainty in the model derives from the uncertainty in the measured photolysis rates. Model sensitivity to input parameters, is discussed in detail in Stone et al. (2010), and show that any overestimation of radical species in the model may be explained by the uncertainty in measured photolysis rates. The model does, however, replicate the variability in observations, as shown in Figure 14 and evidenced by the r 2 correlation coefficient of 0.75 between the modelled and measured HO 2 concentrations.
Rates of production/destruction analyses (ROPAs/RODAs) for the model run using photolysis rates measured on the BAe-146 indicates that the dominant reactions responsible for HO 2 production during the intercomparison periods were OH + CO (39%), CH 3 O + O 2 (15%), HCHO + OH (10%) and HCHO photolysis (9%). While individual contributions to HO 2 production from RO + O 2 reactions are relatively small, they do contribute significantly to the remainder of the total. Dominant loss processes for HO 2 were HO 2 + NO, comprising 35% of the total loss, HO 2 + HO 2 (28%), HO 2 + CH 3 O 2 (11%) and O 3 + HO 2 (8%). Cycling of HO 2 between the HO 2 and HO 2 NO 2 was also a significant process during the intercomparison periods, and reactions of HO 2 with RO 2 species also contributes significantly to the total HO 2 loss.
The dominant peroxy species in terms of HO 2 loss were also the most abundant RO 2 species in the model, with CH 3 O 2 constituting 88% of the total organic peroxy radical concentration. A more detailed RO 2 speciation from the model is given in Table 2.

Summary and conclusions
The measurement of peroxy radicals requires the use of very sensitive and well calibrated measurement techniques. Owing to the particular complexity of airborne measurements, any laboratory characterisation must however be accompanied by in flight intercomparisons and calibrations of the Atmos. Chem. Phys., 10, 10621-10638, 2010 www.atmos-chem-phys.net/10/10621/2010/   instrumentation in order to identify malfunctions under field conditions and to prevent biases in the analysis of results.
In that context, the radical intercomparison exercise within AMMA intends for the first time to compare the performance of a LIF-FAGE, measuring [HO 2 ], and two similar PeRCA instruments, measuring the total sum of peroxy radicals [RO * 2 ], deployed in two different aircrafts and sampling at two pressure levels over the same geographical area.
Some instrumental failures limited the number of data available for comparison. Nevertheless the results obtained provide very valuable information for the analysis of the AMMA campaign as a whole: (a) [RO * 2 -PeRCA 4] and [RO * 2 -DUALER] agree within the instrumental errors when the correction for the expected humidity interference to the chain chemistry in the PeRCA 4 instrument is not applied, while the [RO * 2 -PeRCA 4] corrected data set remains unrealistically higher. The modelled reduction in the PeRCA 4 chain length owing to the pressure and temperature dependencies of the chain chemistry reaction rate coefficients does explain a significant proportion of the systematic offset between the PeRCA 4 and DUALER measurements. The demonstrated lack of a trend in the model to measurement ratio with relative humidity further supports the conclusion that under the conditions of the aircraft measurements the chain length is unperturbed by the ambient levels of humidity. The measurements otherwise show the same temporal variation and agree in magnitude within the instrumental uncertainties. This support the conclusion that it is inappropriate to use the humidity corrected data from the PeRCA 4 instrument for the rest of the AMMA campaign.
(b) The response of the j (NO 2 ) radiometer on board the D-Falcon remains systematically higher than the equivalent radiometer on the BAe-146. This is the result of the 25-35% difference observed in the upward facing sensors at both pressure levels. This difference remains within the combined total errors of both radiometers which is mainly related to the azimuth and zenith dependence of the sensitivity of upward facing radiometers. This emphasises the importance of specific calibrations and verification of 2 π sr radiometers to preventspuriously simulation of radical mixing ratios, as confirmed by the sensitivity study undertaken with the DSMACC model. These discrepancies require exercising caution when comparing model results from the two aircrafts.
Concerning the content of the air masses sampled, the radical mixing ratios vary on the average between 7 and 20 pptv for [HO 2 ] and 15 and 40 pptv for [RO * 2 ] at both pressure levels. The [RO * 2 ] variability increases significantly at the higher pressure level (485 hPa) indicating input of precursors of organic peroxy radicals, in reasonable agreement with other trace gas observations. The existence of a layer photochemically active at this altitude has been reported during AMMA (Andrés-Hernández et al., 2009;Reeves et al., 2010). The [RO * 2 ]/[HO 2 ] ratios vary between 1:1 and 3:1 and seem to be more stable and closer to the expected 2:1 for remote areas at the lower pressure level (697 hPa). According to the laboratory characterisation of the DUALER (Kartal et al., 2010), using a common chain length for the [RO * 2 ] determination during the intercomparison exercise can lead to an 8-16% overestimation of the reported [RO * 2 -DUALER]. The results of this intercomparison campaign confirm the necessity of additional effort to improve not only the inflight cross calibration of airborne measurements for peroxy radicals but also the rapid and high sensitive measurement of photolysis rates and other associated species like VOCs and NO, which play a central role in the discrimination between unexpected experimental malfunctions and unknown atmospheric chemistry. Furthermore, the broad and accurate knowledge of instrumental limitations and potential artefacts in the field is indispensable for the correct use of airborne measurements of short lived species for the modelling of local photochemistry.