The campaign took place between 7 June and 8 July 2014. The measurement site is close to the town of Wangdu (population 260 000 inhabitants), which is without major industry. The Taihang Mountains are located 50 km northwest of Wangdu and the Bohai Sea 200 km east. The next large city, Baoding, is 35 km northeast of Wangdu. Beijing and Shijiangzhuang, two of the largest cities in the North China Plain, are located 170 km northeast and 90 km southwest of the site, respectively. Time is given in this paper as CST (Chinese national Standard Time = UTC + 8 h). Sunrise was at 04:30 CST and sunset at 20:00 CST.

Instruments were set up in a botanical garden, which was surrounded by farmland. Wheat and willows were the dominant plant species, a few of which were growing within 10 m of the instruments. There was no car or truck traffic in the botanical garden; the closest road was 2 km away. Most of the instruments were placed in seven sea containers. Two of them were stacked on top of three others and two more containers were placed approximately 5 m away.

A large number of trace gases and aerosol properties were measured during this campaign. Most of the instrument inlets were placed 7 m above the ground at the height of the upper containers. Table  summarizes the details of the trace gas measurements. OH, HO2 and RO2 radicals were measured by laser-induced fluorescence described in detail below. The OH reactivity (kOH), which is the inverse chemical lifetime of OH, was directly measured by a laser pump and probe technique .

Most of the inorganic trace gases (O3, CO, CO2, NO and NO2) were simultaneously monitored by more than one instrument. Measurements of O3, CO and CO2 measurements agreed well within the instrumental accuracies. O3 measurements were performed by two commercial instruments using ultraviolet (UV) absorption (Environment S.A. model 41M and Thermo Electron model 49i). Also SO2, CO, CO2 concentrations were measured by commercial instruments (Thermo Electron models 43i-TLE, 48i-TLE and 410i). In addition, a cavity ring-down instrument (Picarro model G2401) provided measurements of CO, CO2, CH4 and H2O concentrations.

Chemiluminescence technique was used to detect NO and also NO2 after conversion to NO. Two commercial instruments were deployed by Peking University (PKU) (Thermo Electron model 42i NO-NO2-NOx analyzer), one of which (PKU-PL) was equipped with a home-built photolytic converter, for the detection of NO2, and the other with a catalytic converter (PKU-Mo). The NO2 data from PKU-Mo were not used here, since catalytic converters can cause interferences from other nitrogen–oxygen compounds (e.g., HNO3, PAN). Another instrument was operated by Forschungszentrum Jülich (FZJ) (Eco Physics model TR 780, with a photolytical converter for NO2). Instruments were located in the upper two containers to have inlet lengths as short as possible in order to minimize the correction for shifts in the NO-to-NO2 ratio by the reaction of NO with O3 in the inlet lines. The effect of changes of the NO-to-NO2 ratio by peroxy radicals is negligible due to their small concentrations and their high loss rate in the inlet line. The distance between inlets was less than 5 m. Measurements of the two PKU instruments and the FZJ instrument differed overall by ±20 %, which cannot be explained by their calibration errors. The reason for this discrepancy is not clear. Calibrations of the FZJ instrument were less reproducible (10 %) than in previous deployments, while calibration measurements of the PKU instrument varied only by 1 to 2 %. Fortunately, the calibrations did not show a trend over time, indicating that there was no accumulation of contaminations in the inlet lines. Because of the more stable calibrations of the PKU instruments, the NO and NO2 data used as model input (Sect. 2.4) were taken from the PKU-Mo and PKU-PL instruments, respectively. However, the difference between measurements of different instruments is considered as additional uncertainty in the NO2 and NO measurements.

Six instruments detected HONO using different techniques. Home-built instruments from FZJ and from PKU utilized long-path absorption photometry (LOPAP). In addition, three instruments applied cavity-enhanced absorption spectroscopy (CEAS) for the detection of HONO. They were operated by the US National Oceanic and Atmospheric Administration (NOAA) , by the Anhui Institute of Optics and Fine Mechanics (AIOFM) and by the University of Shanghai for Science and Technology (USST). A gas and aerosol collector (GAC), which is based on the wet denuder/ion chromatography technique, could also detect HONO . The measurements between multiple instruments agreed within 30 %. HONO measurements from the FZJ-LOPAP instrument are used as model constraint, because it showed the best detection limit and temporal coverage during the campaign. Results of model calculations only change less than 10 % if either measurements by the PKU-LOPAP or NOAA-CEAS are instead used as constraint. The other CEAS HONO instruments measured only during a few days. The GAC HONO measurement is known to be affected by interferences from ambient NO2 and was therefore not used here.

A total of 59 organic species were measured by a gas chromatograph (GC) equipped with a mass spectrometer and a flame ionization detector (FID) . This instrument provided concentrations of C2 to C11 alkanes, C2 to C6 alkenes, and C6 to C10 aromatics. In addition, measurements of VOCs were performed by a proton transfer reaction – mass spectroscopy system (PTR-MS, Ionicon). These measurements included isoprene, acetaldehyde, the sum of methyl vinyl ketone (MVK) and methacrolein (MACR), benzene, toluene, styrene, C8 aromatics, C9 aromatics and acetonitrile. Daytime measurements of the two instruments agreed well for those species which were detected by both instruments. During nighttime, however, PTR-MS measurements gave much larger values compared to measurements by GC for some periods and some species. The reason for that is not clear, but could have been caused by interferences by other species that occur at the same mass in the PTR-MS. Because of this uncertainty, mainly measurements by GC are taken as constraints for model calculation here. Measurements of acetaldehyde, MVK and MACR were only done by PTR-MS. Formaldehyde (HCHO) was measured by a commercial instrument utilizing the Hantzsch method (Aerolaser GmbH model AL4021).

A 20 m high tower with meteorological instrumentation was set up 15 m south of the containers, where temperature, pressure, relative humidity, wind speed and wind direction were measured at two different heights (10 and 20 m). The height of the planetary boundary layer (PBL) could be estimated by a ceilometer (the minimum detectable PBL height was 200 m). Photolysis frequencies were calculated from the spectral actinic photon flux density measured by a spectroradiometer , whose inlet dome was placed on top of the highest container.

More trace gases were detected, but will not be discussed in detail here; peroxyacyl nitrates (PAN) and peroxypropionyl nitrate (PPN) were measured by gas chromatography with an electron-capture detector . H2O2 was collected by a scrubbing coil collector and detected by high-performance liquid chromatography (HPLC) coupled with post-column derivatization and fluorescence detection . Chemical ionization mass spectroscopy (CIMS) was utilized to measure nitryl chloride (ClNO2) and N2O5, but measurements were only conducted after 21 June . A cavity-enhanced absorption spectrometry instrument was deployed to detect glyoxal, HONO and NO2 .

Aerosol properties were characterized in detail during the campaign, but will be discussed elsewhere. Measurements included particle number density and size distribution and also chemical composition.

A box model is used to simulate the concentrations of OH, HO2, RO2 and RO2# and the total OH reactivity. The model is based on the compact Regional Atmospheric Chemical Mechanism version 2 (RACM) described in . This mechanisms includes 17 stable inorganic species, 4 inorganic intermediates, 55 stable organic compounds and 43 intermediate organic compounds. Compounds that are not explicitly treated in the RACM are lumped into species with similar functional groups. The assignment of organic compounds that were measured during this campaign to species in the RACM is listed in Table .

Some modifications were applied to the RACM. The isoprene mechanism was replaced by the more detailed mechanism listed in Table . It is based on the Leuven isoprene mechanism (LIM) proposed by . Here, we use the updated LIM for bulk RO2 reactions described in . In addition, the chemistry of the first-generation products of the isoprene oxidation, MVK and MACR and isoprene hydroperoxides (ISHP), are revised. MACR has been shown to regenerate OH via RO2 isomerization and decomposition . OH is also formed by the reaction of RO2 from MVK with HO2 with a significant yield . The products of the reaction of isoprene hydroperoxides formed in the reaction of isoprene RO2 with HO2 have been revised by , showing that epoxides can be formed in an OH neutral reaction. The modified RACM 2 in this work has been compared to the modified RACM-MIM-GK which was used previously for model studies of the HOx chemistry in China . In the present study, modeled HOx concentrations differ no more than 5 % between the old and new modified RACM mechanisms. It is also noteworthy that HOx results of the modified RACM-MIM-GK agreed well with predictions of the more explicit Master Chemical Mechanism v3.2 .

Model calculations are constrained to measured trace gases, including inorganic species (H2O, NO, NO2, O3, HONO, CO) and organic species (methane and nonmethane organic compounds listed in Table ). Because only the sum of MVK and MACR were measured, a ratio of 0.6 : 0.4 was used to divide the sum measurement to individual species. In addition, physical parameters like photolysis frequencies, temperature and pressure are constrained to measured values.

For model calculations, the measured time series are synchronized to 5 min time intervals. This is done either by averaging or by linear interpolation, if the time resolution of the measurement is shorter or longer than 5 min, respectively. Measurements of the two instruments for ozone and CO are combined in order to fill data gaps.

Slightly more than 60 % of the measured OH reactivity can be explained by the measured concentrations of CO, NOx and hydrocarbons during daytime. More than 90 % of the OH reactivity can be explained if also measured oxygenated VOC species are included . Consequently, there were no large amounts of other relevant OH reactants in the atmosphere which would otherwise have contributed significantly to the measured reactivity. For this reason, long-lived product species which were not measured are constrained to zero in the model, in order to avoid unrealistic build-up of additional reactivity. This constraint is consistent with the assumption that most of the measured pollutants were emitted nearby and were not photochemically aged. Only aldehydes (ALD) are not set to zero, because they lead to the formation of reservoir species for organic peroxy radicals (peroxy acyl nitrates, PAN and PPN), which are kept as free parameters. In addition, HPALD that is formed in the new isoprene chemistry is not constrained to zero. In order to avoid unrealistic accumulation of oxygenated VOC species (mostly aldehydes), an artificial, constant loss is added, which limits their lifetime to 24 h.

For comparison with experimental data, the modeled concentrations of individual RO2 species are summed up in two categories which simulate the measured total RO2 and RO2# concentrations (cf. Sect. 2.3.4). Modeled RO2 contains those species that can be detected by the measurement system. The largest class of RO2 that is not included in the calculated RO2 are NO3-alkene adducts (RACM name OLND), because their reaction with NO does not produce HO2. The largest concentration of OLND is predicted in the early evening (approximately 1×108 cm-3). In contrast, the majority of modeled RO2 during daytime consists of species which are detected. In the model, the observable RO2 species contribute with equal weight to the total RO2, whereas laboratory calibrations of the RO2 instrument have shown slightly different (less than ±20 %) detection sensitivities for the measured RO2 species . Modeled RO2# represents a subclass of RO2 species which are produced in RACM 2 from alkenes, aromatics and long-chain (> C4) alkanes.

The relatively large uncertainty of the model calculations is a combination of uncertainties in the measurements used as model constraints and reaction rate constants (for details, see ). Differences in the measurements of NO (20 %) and HONO (30 %) from different instruments change modeled OH concentrations by only 7 and 10 %, if measurements from one or the other instrument is taken as constraint. The uncertainties of measurements and modeling need to be taken into account in the comparison. The uncertainty of radical measurement is mainly determined by the 1σ measurement accuracies (OH: ±11 %, HO2: ±16 %, RO2: ±18 %). A series of tests based on Monte Carlo simulations show that the uncertainty of the model calculations is approximately 40 %.

Chemical modulation tests as described in Sect. 2.3.2 were conducted on 29 June (afternoon), 30 June (morning and afternoon), 2 July (afternoon) and 5 July (afternoon and evening). The time periods of the tests and the atmospheric chemical conditions are given in Table . All test results are shown in Fig. , where the measured OH signal SN2 (without OH scavenger) is compared to the sum of the expected signals from ambient OH (SOH) and the known O3 interference (SO3). Statistical error bars shown in Fig.  are derived from 1σ measurement precisions of SN2 and Sprop. In addition, the sum of SOH and SO3 has a systematic error (not shown in Fig. ), which is dominated by the uncertainty (±10 %, 1σ) of the removal efficiency (ϵ) needed to calculate SOH (see Eq. ).

The signals SN2 (Fig. ) are on average higher than the corresponding sum of SOH and SO3. The differences vary within the range between 0.53 ×106 and 1.2 ×106 cm-3 (Table ) and could be the result of an unknown OH interference or of the systematic experimental error in the determination of SOH + SO3. The differences are subject not only to statistical errors, which are shown as the error bars in Fig. , but also to the uncertainty arising from the calculation of SOH (Eq. ). Among all, the uncertainty in the removal efficiency (ϵ) has the largest impact on the derived differences. The differences between SN2 and SOH + SO3 and their uncertainty are listed in Table . No correlation of differences with time of day or with the chemical conditions is observed. The differences fall quantitatively into the 2σ range of the accuracy of SOH + SO3 and are therefore at the limit of detection of the experimental setup used in the campaign. Because the test results are not sufficiently accurate to draw firm conclusions about an unknown interference, the OH data in this work were not corrected for a potential interference. Instead, the differences found in Fig.  are treated as an additional uncertainty of the OH measurements presented in this paper.

In the case of an interference, it would be a small fraction of the measured OH during daytime. The measured nighttime OH, however, would be much more affected. Because the existence of an unknown OH interference cannot be strictly ruled out, the interpretation of the radical chemistry will therefore concentrate on daytime conditions. More precise and accurate chemical modulation tests with an improved experimental setup are needed in future field campaigns.

Meteorological conditions were characterized by high temperatures of up to 37 ∘C and high humidity. The wind velocity was usually below 2 m s-1. Back trajectory analysis using the NOAA HYSPLIT (Hybrid Single Particle Lagrangian Integrated Trajectory Model) model showed that air masses were often transported from south or east where large city clusters are located. Solar radiation was strong during this campaign with few exceptions of hazy or cloudy days (15–19, 25 June and 1 to 4 July; Fig. ).

Afternoon CO mixing ratios increased during several periods, indicating accumulation of anthropogenic emissions on a regional scale. They are separated by sudden drops during rain events on 19 June and 4 July and on 27 and 28 June when clean air was transported from the north.

During the first half of the campaign, burning of agricultural waste after harvesting in surrounding fields was observed. This was confirmed by high acetonitrile mixing ratios (> 1 ppbv) from 12 to 19 June. Biomass burning was accompanied by a reduced visibility and an increase in aerosol mass concentrations (PM2.5) with maximum values of 150 µg cm-3 on 16 June (campaign average value: 70 µg cm-3).

Time series of O3 and NO2 often showed trends similar to CO, but were also strongly influenced by photochemistry. Maximum daily ozone mixing ratios ranged between 100 and 140 ppbv depending on the strength of radiation. Because solar radiation was attenuated between 14 and 19 June during the first pollution episode, O3 peaked already on 14 June. O3 was sometimes completely titrated by nitric oxide at night.

Isoprene mixing ratios exhibited a typical diurnal profile with maximum values between a few hundred pptv and nearly 4 ppbv in the afternoon. These values indicate that chemical conditions were also influenced by presumably local biogenic emissions.

The time series of measured and modeled OH, HO2, RO2 and kOH are shown in Fig. . Distinct diurnal profiles are observed for all radical species. The daily maxima of OH, HO2 and RO2 appeared around noontime and concentrations ranged between (5–15) ×106 cm-3, (3–14) ×108 cm-3 and (3–15) ×108 cm-3, respectively. On 18, 19 and 25 June and from 1 to 3 July, radical concentrations were low due to attenuated solar radiation. On 28 June, OH increased to exceptionally high concentrations of up to 3×107 cm-3 for a short period of time, which was accompanied by an increase of the HONO mixing ratio to 2 ppbv, leading to enhanced OH production from HONO photolysis. During this time, farmland next to the measurement site was treated with water and artificial nitrogen-containing fertilizer, which may have caused large local HONO emissions.

In general, the model reproduces the measured time series of OH, HO2 and RO2 well. Differences between modeled and measured radical concentrations are generally smaller than the combined 1σ uncertainties of radical measurements (10 %) and model calculations (40 %).

A closer look at the modeled and measured radical concentrations reveals some systematic trends. Modeled OH concentrations tend to be smaller than measurements during afternoon hours and modeled RO2 concentrations tend to be lower in the early morning and higher in the evening than corresponding RO2 measurements. In contrast, differences between modeled and measured HO2 concentrations are small at all times. Because of the similarity of the model–measurement agreement for different days, further analysis of daytime radical concentrations will be done on the basis of median diurnal profiles (Sect. 3.4).

The OH observed at night are mostly above the limit of detection (3 ×105 cm-3) with concentrations around 5 ×105 cm-3, whereas the model predicts concentrations below the limit of detection. In a few nights, the measured OH is even higher (e.g., 1–3 ×106 cm-3 on 13 June). The reason why the measured OH values are significantly higher than the model prediction is not clear. It could be caused by missing chemistry in the model or vertical gradients in the nocturnal boundary layer, as discussed in . Furthermore, we cannot exclude an unknown interference of the same magnitude. The known interference from NO3 is probably not sufficient as an explanation ; the expected interference would be 1 ×105 cm-3 for this campaign, which is 5 times less than the averaged nighttime OH measurement.

Thus, if interferences played a role, they would probably have a different origin.

The time series of measured OH reactivity shows a change on 20 June (Fig. ). During the first 2 weeks, diurnal profiles of kOH are more structured and show higher values with maximum values of up to 40 s-1 compared to values after 20 June, when kOH is only around 10 s-1 in the afternoon and exhibits a less distinct diurnal profile. The first period coincides with the accumulation of pollutants like CO, nitrogen oxides and particles (Fig. ). In addition, harvesting and biomass burning activities caused local emissions of OH reactants, which may explain the short-term increases in OH reactivity during this period, especially during nighttime, when fresh emissions are released into the shallow nocturnal boundary layer and highest OH reactivity is observed. After 20 June, biomass activities close to the measurement place were less often observed and heavy rainfall cleaned the air.

In the first period of the campaign, the model often underpredicts the measured OH reactivity, especially at night. This is likely caused by unmeasured atmospheric compounds from local emission sources like biomass burning. In the second period, the modeled and measured reactivities agree well during day and night for most of the time.

Differences between measurements and model calculations are further analyzed using median diurnal profiles with a time resolution of 1 h (Fig. ). Data are only included when measurements of all key species used as model constraints and radical measurements are available at the same time. Therefore, 4 days are excluded from the analysis from the entire data set. On 13 June, data gaps are larger than 6 h for nearly all instruments. No measurements of VOCs are available on 14 June, no measurements of photolysis frequencies on 22 June and no radical measurements on 4 July. As described in Sect. 3.3, chemical conditions were slightly different before and after 20 June. We found similar results of model–measurement comparisons for radicals from the two periods for daytime conditions. Therefore, the following interpretation and discussion will focus on campaign-averaged diurnal profiles. Chemical conditions of data included in the median profile are summarized in Table  and median diurnal profiles of important photochemical parameters are shown in Fig. .

The median diurnal profiles of the measured and modeled OH concentrations agree within their errors of 10 % (1σ) and 40 %, respectively, from sunrise to mid-afternoon. When the median NO mixing ratio (cf. Fig. ) drops gradually from 0.3 ppbv to 0.1 ppbv in the afternoon, a systematic difference evolves, with measured OH concentrations being approximately 1 ×106 cm-3 higher than the model calculations. The discrepancy is of similar magnitude to the averaged unexplained OH determined in the chemical modulation experiments (Table ). Thus, the overall agreement for OH would improve if the unaccounted signal was fully considered as an OH measurement interference. However, the underestimation of OH would persist for low NO conditions if a potential unaccounted signal was subtracted. When NO concentrations are less than 100 pptv, the observed-to-modeled OH ratio would be reduced from 1.9 to 1.5, indicating that an OH source would still be missing for low NO conditions. Although newly proposed isoprene mechanisms have the potential to enhance the OH regeneration for low NOx conditions, they only have a small effect on modeled OH concentration at the conditions of this study with NO concentrations higher than 0.1 ppbv and isoprene concentrations lower than 2 ppbv.

In general, HO2 concentrations are reproduced by the model during daytime within the combined uncertainties of measurements and model calculations. Nevertheless, the model has a tendency to overpredict HO2 in the afternoon. If we constrain the model to the observed HO2 concentrations, the observed-to-modeled OH ratio increases from 1.6 to 1.8 for daytime-averaged conditions (04:30–20:00 CST). RO2 and RO2# are significantly underestimated during the morning hours (06:00–10:00 CST) with an observed-to-modeled ratio of 3 to 5, which is larger than the combined uncertainty (a factor of 2). Reasons for discrepancies between measured and modeled RO2 are further analyzed in Sect. 3.6.

Measured kOH is high at night, peaks in the morning (22 s-1) and decreases to about 11 s-1 in the afternoon. Modeled kOH shows a relative flat diurnal profile (average over the day is 14 s-1). Whereas good agreement with measurements is achieved during daytime, measured reactivity is higher during nighttime especially during the first part of the campaign. This is likely caused by unmeasured emitted OH reactants. A sensitivity model run, in which product species are not constrained to zero as in this model run, does not give significantly different OH reactivity in the night. A more detailed analysis of the OH reactivity in this campaign is presented in our companion paper by .

Strong correlation has been found between j(O1D) and OH radical concentrations for many field campaigns in different environments from marine to continental locations . A strong linear correlation is also observed for data from this campaign (Fig. ). A linear fit between measured OH concentrations and measured photolysis frequencies yields a slope of 4.5 ×1011 scm-3. This value is similar to values that were derived in previous field campaigns in China in 2006 in the Pearl River delta and Yufa .

The intercept of the linear fit for the campaign in Wangdu is 1.0 ×106 cm-3, which is smaller than intercepts obtained for the data set from the campaigns in the Pearl River delta (2.4 ×106 cm-3, ) and Yufa (1.6 ×106 cm-3, ). The intercept gives an estimate of the importance of radical sources when the production of O1D from ozone is small. This includes non-photolytic sources (e.g., ozonolysis of VOCs) and photolytic processes in the early morning before j(O1D) starts to rise (Fig. ).

Modeled OH also shows a strong dependence on j(O1D) with a slightly smaller intercept compared to the fit result using the measurements.

Figure  shows median diurnal profiles of measured RO2 and RO2# together with modeled concentrations of speciated RO2 radicals. The observed profiles of RO2 and RO2# have similar shapes with a maximum around 14:00 CST. In the morning hours, RO2 is dominated by RO2#, whereas RO2# plays only a minor role in the late afternoon and at night. The model reproduces the general behavior of RO2 and RO2# well, with very good agreement in the afternoon. However, in the morning, the model underestimates RO2 systematically by a significant amount of (1–2) ×108 cm-3. This is mainly caused by an underestimation of RO2#. After sunset, in the first half of the night, the model overestimates RO2. This discrepancy is apparently related to organic peroxy radicals, which do not belong to RO2#.

In the group of modeled RO2# species, isoprene peroxy radicals (ISOP) make the largest contribution during daytime. Other modeled RO2# include peroxy radicals from alkenes, aromatics, long-chain (> C4) hydrocarbons, and MVK and MACR. Among the RO2 radicals which do not belong to the RO2# group, peroxy radicals of short-chain (< C5) alkanes are dominating: methyl peroxy radicals (MO2), ethyl peroxy radicals (ETHP) and peroxy radicals from HC3P (e.g., propane). Acetyl peroxy radicals (ACO3 + RCO3) are also a substantial fraction of RO2.

The strong underprediction of the observed RO2 by more than a factor of 4 in the morning cannot be explained by the measurement errors and interferences discussed in Sect. 2.3.4 and 2.3.5. In order to explore potential reasons for this underprediction, several sensitivity tests were performed. First, the impact of a faster OH on RO2 conversion by an increased amount of VOC was tested (model sensitivity run S1). Second, an additional primary source of RO2 was introduced into the chemical mechanism (S2). Third, the possibility of a slower removal rate of RO2 was tested (S3).

The first possibility (S1) is supported by the observation that the modeled OH reactivity in the base run (S0) is smaller than the measured OH reactivity in the morning until about 09:00 CST. If this missing reactivity is caused by unmeasured VOCs, the true RO2 production from reactions of VOCs with OH would be larger than the modeled one. To fill this gap, the total concentration of the measured VOCs is increased to match the measured kOH in the time window from 06:00 to 09:00 CST. The relative partitioning of the VOCs is not changed. The model run (S1) with the upscaled VOC reactivity resolves part of the RO2 discrepancy until 09:00 CST (Fig. ). The observed-to-modeled RO2 ratio is improved from 2.8 to 1.7 without affecting the good model–measurement agreement for OH and HO2. Further sensitivity tests show that the modeled RO2 is not sensitive to the speciation of the additional VOC reactivity, since the required change of kOH is relatively small (< 20 %). Because no missing OH reactivity is found after 09:00 CST, the gap between measured and observed RO2 cannot be explained by unmeasured VOCs later.

In sensitivity test S2, an additional primary source of RO2 (OLTP) from terminal alkenes is introduced into the model. A source strength of 2 ppbvh-1 from 06:00 to 12:00 CST would be required to achieve a good model–measurement agreement (within 20 %) for both RO2 and RO2#. The modeled OH and HO2 concentrations also increase and are slightly overpredicted by about 10 and 20 %, respectively. This can still be considered as agreement within the error of measurements and model calculations. After 12:00 CST, the difference between modeled and measured RO2 becomes smaller than 15 %, within the range of the accuracy of RO2 measurements.

A candidate for an additional primary RO2 source would be reactions of VOCs with chlorine atoms, which are produced by photolysis of nitryl chloride (ClNO2) . ClNO2 is formed from the heterogeneous reactions of Cl- ions with nitrogen pentoxide (N2O5) and accumulates during nighttime. After sunrise, ClNO2 is expected to be completely photolyzed within a few hours. The resulting Cl atoms can abstract H atoms from saturated hydrocarbons or can add to alkenes. The alkyl radicals produce RO2 which, in the case of alkene-derived peroxy radicals, carry a chlorine atom. ClNO2 was measured by a CIMS instrument at the Wangdu field site from 20 June to 8 July . The concentrations increased at night and reached on average high values of 0.5 ppbv at 08:00 CST, followed by a decay to zero until 11:00 CST. In their study, investigated the role of ClNO2 photolysis on the photochemical formation of RO2 and ozone during the Wangdu campaign. They used the Master Chemical Mechanism (MCM) v3.3 with an additional chlorine chemistry module by . We repeated the study by adding the same chlorine chemistry to our modified RACM 2 mechanism and found the same additional formation rates of RO2 and O3 as reported by . In our model run, a ClNO2 source is assumed that leads to a linear increase of ClNO2 concentrations during nighttime to a maximum value of 0.5 ppbv at 08:00 CST for every day. After 08:00 CST, the modeled source is turned off. ClNO2 starts to photolyze after 06:00 CST with a photolysis frequency that was calculated from the measured actinic flux. A maximum Cl production rate of 0.2 ppbvh-1 is obtained at 08:00 CST, yielding an additional RO2 production with a similar rate. Compared to the additional RO2 production rate required for model run S2, this is an order of magnitude too small. The mechanism is also not capable of sustaining the additional RO2 production during the whole morning, because ClNO2 is photolytically depleted within 2–3 hours. Even if the modeled source strength is increased to match the highest ClNO2 mixing ratio of 2 ppbv observed on 21 June , the additional primary RO2 production of 0.5 ppbvh-1 is still not sufficient. Thus, although ClNO2 photolysis was a relevant radical source, it alone cannot explain the missing source of RO2 radicals in the morning.