The University of Birmingham operated a LOPAP (QUMA Elektronik & Analytik GmbH) at IAP. The LOPAP is a wet chemical technique and has been described in detail in Heland et al. (2001) and Kleffmann et al. (2002). Briefly, a stripping coil is used to sample gas-phase HONO into an acidic solution where it is derivatized into an azo dye. The light absorption of the azo dye, principally at 550 nm (though higher wavelengths can also be used), is then measured with a spectrometer using an optical path length of 2.4 m. The LOPAP was operated and calibrated according to the standard procedures described in Kleffmann et al. (2008). The time resolution of the LOPAP was 5 min and baseline measurements were taken at frequent intervals (8 h). The operationally defined detection limit (2σ) of the LOPAP was calculated to be 35 and 5 ppt for winter and summer, respectively and varied due to changes in purity of reagents and zero air used. The LOPAP was housed within a temperature controlled shipping container and sampled at a height of 3 m above ground level.

Institute of Chemistry, Chinese Academy of Sciences (ICCAS) applied a custom-built instrument, described in detail elsewhere (Hou et al., 2016). It is a wet chemical technique similar in principle to the LOPAP. Gas-phase HONO is almost completely absorbed by an absorption solution into a two-channel stripping coil, where it forms an azo dye, detected by absorption spectroscopy at a wavelength of 550 nm with an optical path length of 0.5 m. The instrument has a detection limit (2σ) of 134 ppt for a response time of 5 min. The ICCAS and BHAM instruments both used a similar outdoor sampling unit that employed a short quartz inlet (< 2.5 cm). While the BHAM and ICCAS instruments operated according to the same principles, there were two main differences. The first was the method for determining the baseline, the BHAM instrument used an overflow of N2 while the ICCAS instrument replaced the reagents with water. The second was the optical path length, which was 2.0 and 0.5 m for the BHAM and ICCAS instruments, respectively.

The University of Cambridge ran a three-channel BBCEAS instrument during the campaign, with one channel measuring NO2 and HONO simultaneously in the UV (362–374 nm) wavelength region. Reference absorption cross sections of HONO (Stutz et al., 2000) and NO2 (Voigt et al., 2002) were fitted to the absorption coefficient to retrieve HONO and NO2 concentrations. Details of the instrument can be found in Kennedy et al. (2011). Two mirrors (Layertec 109053) with peak reflectivity of ∼99.95 % at 365 nm were used to form the cavity. Given an inter-mirror distance of 92 cm, the effective absorption pathlength in the case of an empty cavity was around 1.8 km. Both the instrument inlet and the optical cavity were made of perfluoroalkoxy alkane (PFA) which is well known for its chemical inertness. The inlet line was 1/4 in. outer diameter PFA tubing and was approximately 3 m long. During the winter phase of the campaign, instrument inlet was placed at the top of the container and was about 3 m from the ground. During the summer phase however, the instrument was moved to an adjacent container, also housing the other BBCEAS instrument, and the height of instrument inlet was changed accordingly from ∼3 to ∼2 m.

To allow a more stable cavity throughput (i.e. to minimize flow turbulence effect on the optical signal), the sampling flow was set to 2 L min-1 for the HONO/NO2 measurement channel. This was close to the very low end of the operational range of the flow controller (0–50 L min-1), leading to the actual flow rate potentially differing from that set. Post-campaign analysis identified that this affected both the dilution factor (dilution of the sample flow by the two mirror purge lines) and the length the sample gas occupied the cavity. A post-campaign calibration was therefore performed by injecting a known amount of NO2 into the cavity under otherwise identical operating conditions, and a scaling factor of ∼1.27 was found to be necessary to account for these two factors and was then applied to the measured NO2 and HONO concentrations.

The custom-built BBCEAS instrument from the Anhui Institute of Optics and Fine Mechanics (AIOFM), Chinese Academy of Sciences, has been described in detail in Duan et al. (2018); therefore only a brief description is given here. Light is emitted by a single light-emitting diode (LED) with peak wavelength of 365 nm, full width at half maximum (FWHM) of 13 nm and is introduced into the resonant cavity, consisting of a pair of high-reflective (HR) mirrors with reflectivity of about 0.99985 at 368 nm, separated by 55.0 cm. The emergent light intensity passing through the cavity exit mirror is received by an Ocean Optics QE65000 spectrometer through an optical fibre with 600 µm diameter and a 0.22 numerical aperture.

In order to avoid the drift of the centre wavelength of the LED, the temperature of the LED was controlled to be approximately 20±0.05 ∘C by using a thermoelectric cooler (TEC) unit. In order to prevent particulate matter from entering the cavity and reducing the effect of particulate matter on the effective absorption path, a 1 µm PTFE filter membrane (Tisch Scientific) was used in the front end of the sampling port. The time resolution of the BBCEAS instrument was 1 min, and the 2σ detection limit of HONO was about 120 pptv. The fitting wavelength range was selected as 359–387 nm, with the same reference cross sections used in the retrieval of NO2 and HONO as for the University of Cambridge instrument. Sample loss and secondary formation of HONO were both considered in this instrument and the measurement error of HONO was estimated to be approximately 9 %. The inlet line was 1/4 in. outer diameter PFA tubing and was approximately 4 m long.

A time-of-fight chemical ionization mass spectrometer (ToF-CIMS) (Lee et al., 2014), using an iodide ionization system was coupled with a filter inlet for gases and aerosols (FIGAERO) originally developed by Lopez-Hilfiker et al. (2014) and recently described and characterized by Bannan et al. (2019). The detailed setup during this campaign can be found in Zhou et al. (2018). The FIGAERO enabled near simultaneous, real-time measurements of both the gas and particle phase composition. Only gas-phase data are presented here, so with every 75 min of continuous data 35 min (particle phase mode) are omitted. The gas-phase inlet consisted of 5 m 1/4 in. I.D. PFA tubing connected to a fast inlet pump with a total flow rate of 13 slpm (standard litres per minute) from which the ToF-CIMS sub-sampled 2 slpm.

Methyl iodide gas mixtures (CH3I) in N2 were made up in the field using a custom-made manifold (Bannan et al., 2014). A total of 20 sccm (standard cubic centimetres per minute) of the CH3I mixture was diluted in 4 slpm N2 and ionized by flowing through a Tofwerk X-ray ionization source. This flow enters an ion molecule region (IMR), which was maintained at a pressure of 400 mbar using an SSH-112 pump fitted with an Aerodyne pressure control box to account for changes in ambient pressure. The IMR pressure is significantly higher than is usual for this CIMS instrument when using Po-210 but is necessary given the change in ionization source in this study. Operation is comparable to the Le Breton et al. (2018) study, who also used the same Tofwerk X-ray ionization source. A short segmented quadrupole (SSQ) was positioned behind the IMR and was held at a pressure of 2 mbar using a Tri scroll 600 pump.

The CIMS instrument zero was determined by flowing dry nitrogen into the IMR periodically, and the backgrounds were applied consecutively. As shown in Fig. S1 (Supplement), there was very little variability of this background during the measurement period. Though the overflowing of dry N2 will have an effect on the sensitivity of the instrument to those compounds whose detection is water dependent, due to the low instrument backgrounds, the absolute error remains small, and we deem this an acceptable limitation in order to measure a vast suite of different compounds for which no best practice backgrounding method has been established. We therefore calculated the absolute error of 33 ppt as 3σ deviations of the background signal.

Field calibrations were regularly carried out using known concentration formic acid gas mixtures made in the manifold. The instrument was calibrated for a range of other species after the campaign, and relative calibration factors were derived using the measured formic acid sensitivity as has been performed previously (Le Breton et al., 2014, 2017; Bannan et al., 2014, 2015). HONO was measured at m/z174 as I.HNO2- during the period of 27 May–17 June 2017. A stable and pure gas-phase source of HONO was generated for calibrations using the method described by Ren et al. (2010) and Febo et al. (1995), and a sensitivity of 0.28 cps ppt-1 was applied to the data with a limit of detection (LOD) of 33 ppt. Data analysis is performed using the “Tofware” package (version 2.5.11) running in the Igor Pro (WaveMetrics, OR, USA) environment. The mass axis was calibrated using I-, I2- and I3-. Extracted high-resolution time series were then normalized to the iodide reagent ion trace. A limitation of the CIMS calibration approach for HONO is that it was not established as a function of humidity. This was not deemed necessary because there was an average variation of only 2 % in the I-:IH2O- ratio throughout the day.

The data presented in this paper has been measured using a Voice200 Selected ion flow tube mass spectrometer (SIFT-MS, Syft Technologies, Christchurch, New Zealand). This instrument consists of a switchable reagent ion source capable of rapidly switching between multiple reagent ions. The ion source region, where the reagent ions are generated in a microwave discharge, acts on an air–water mix at a pressure of approximately 440 mTorr (1 mTorr =0.133 Pa) to generate the three reagent ions H3O+, NO+ and O2+. These ions are extracted into the upstream quadrupole chamber maintained at a pressure of approximately 5×10-4 Torr, using a 70 L s-1 turbo-molecular pump. The reagent ions pass through an array of electrostatic lenses and the upstream quadrupole mass filter, and those not rejected by the mass filter are passed into the flow tube where they are carried along in a stream of nitrogen and selectively ionize target analytes. Gas-phase data presented herein were determined using the H3O+ reagent ion only. Sampling was carried out at a height of ∼3 m using a gas-phase inlet consisting of 3.5 m 1/4 in. I.D. PFA tubing connected to a diaphragm inlet pump (KNF) at a total flow rate of 5 slpm, from which the SIFT-MS sampled approximately 2 slpm through an in-house-built pressure-controlled inlet maintaining a consistent absolute inlet pressure of 0.5 bar. The flow tube is pumped by a 35 m3 h-1 scroll-type dry pump (Edwards) resulting in a mass flow controlled gas flow of 25 sccm for the nitrogen carrier gas (research grade, BOC) and a sample flow of 100 sccm from the pressure controlled inlet system. These flows result in a continuous total flow tube pressure of 460 mTorr and a reaction time of approximately 8 ms (Hera et al., 2018). During the campaign, gas-phase backgrounds were established through regularly overflowing the sample inlet with dry nitrogen for 5 continuous minutes every hour. The determined HONO gas-phase backgrounds in nitrogen were 110±40 pptv during the measurement period presented, and as such are unlikely to have a significant contribution on the ambient mixing ratio.

The bimolecular reaction of H3O+ and nitrous acid produces the product ions H2NO2 (m/z48, 67 %) and NO+ (m/z30, 33 %). The rate constant (k) of this exothermic proton transfer reaction is calculated to be 2.7×10-9 cm3 s-1 with respect to hydronium (H3O+) and 2.2×10-9 cm3 s-1 with respect to hydronium mono-hydrate (H3O⋅H2O)+ (Spanel and Smith, 2000). Nitrous acid does not undergo proton transfer with hydronium di-hydrate (H3O⋅H2O2)+ and tri-hydrate (H3O⋅H2O3)+ in SIFT-MS. Nitrous acid mixing ratios herein were determined using the branching-ratio-corrected protonated product ion m/z48 intensity normalized to both H3O+ and H3O⋅H2O with their respective k values (Taipale et al., 2008). As such, calculated HONO mixing ratios using SIFT-MS should be independent of the humidity of the gas sample.

The time series (Fig. 1) demonstrates that while all instruments captured the same temporal trends, the absolute concentrations differed. The correlation coefficients from regression analyses show that there is little scatter between measurements from the different instruments with r values being consistently between 0.96 and 0.98 (Table 2 and Fig. S4). Overall, the BHAM LOPAP measurements were consistently the highest, followed by ICCAS, AIOFM and CAM. The slopes from the RMA analysis demonstrated that none of the instruments were in agreement (Table 2) within their stated error (Table 1) during the formal intercomparison exercise. Therefore, in the following sections we investigate possible reasons to account for the lack of agreement between instruments.

We calculated the coefficient of variance (CV) as a measure of the precision between the four instruments as per Eq. (1): 1CV=σμ, where μ is the mean and σ is the standard deviation for the measurements by all four instruments at a given 5 min interval. The CV was used to compare the relative degree of variation between datasets and as a guide a CV of 0.1 is considered as acceptable by the US EPA for particulate matter PM instruments (Sousan et al., 2016). From Fig. 2, the CV was fairly consistent throughout the winter intercomparison, at an average of 0.28±0.07. The CV was however observed to increase at the end of the intercomparison, coinciding with period of the lowest mean HONO concentration (< 1 ppb, Fig. 2). An increase in the CV indicates worsening agreement between instruments, possibly due to the concentrations approaching the detection limit (DL) of some instruments (Table 1) NO2 is a potential interferent for the measurement of HONO for both wet chemical and BBCEAS instruments (Heland et al., 2001). Both BBCEAS instruments use the Voigt et al. (2002). NO2 cross section, which has previously been shown to have negligible HONO absorption structures (Veitel, 2002; Kleffmann et al., 2006). We also note that HONO reference spectra should contain little structure from NO2. Overall, Fig. 2 demonstrates no apparent relationship between the CV and NO2. This likely reflects the efforts taken during processing and measurement to reduce the influence of interference from NO2 in all instrument types.

Firstly, the systematic error for each instrument can be calculated by normalized sequential difference (NSD) according to Eq. (2) (Arnold et al., 2007): 2NSD=(Conct-Conct+1)(Conct×Conct+1)0.5. NSD is a method of calculating the variation between consecutive measurements for an individual instrument, where Conct is the concentration measured at time t and Conct+1 the following measurement. The results are shown in Fig. S5, and as each instrument showed a symmetrical and Gaussian distribution it suggests there was no internal systematic bias for any given instrument.

Secondly, we then examined the normalized difference (ND) between pairs of instruments to explore inter-instrument variability, calculated according to Eq. (3) (Pinto et al., 2014): 3NDij=(Ci-Cj)(Ci+Cj), where Ci and Cj denote HONO levels measured by any pair of instruments (BHAM, ICCAS, AIOFM or CAM) calculated for each measurement period. For example, the ND for the BHAM and CAM instruments (NDBHAM-CAM) would be calculated by ([HONO]BHAM- [HONO]CAM)/([HONO]BHAM+ [HONO]CAM). We also calculated the coefficient of divergence (CD), which is a normalized measure of the similarity between two measurement time series, derived via Eq. (4) (See Pinto et al. (2014) and references therein): 4CDij=1/p×∑NDij2, where p is the number of observations and NDij is defined in Eq. (3). A CD of 1 means the time series are completely different, while of CD of 0 indicates that they are identical. The calculated CD for each instrument pair is shown in Table 3 and demonstrates that each of the two overall approaches – wet chemical (BHAM and ICCAS) and BBCEAS (AIOFM and CAM) – agreed well internally. The ICCAS and AIOFM also agreed well, but CAM and BHAM had a higher CD with AIOFM and ICCAS (Table 3).

If there is no difference between a pair of instruments, then the calculated ND should be scattered around 0, and from Fig. 3 this was not observed for any instrument pair, pointing to differences between instruments. The ND was evaluated as function of wind direction and measured HONO concentration (Fig. 3) to explore if ambient concentration or spatial heterogeneity could explain the disagreements. From Fig. 3, for all instrument pairs, the highest ND, and therefore largest relative difference between instruments, was at low HONO mixing ratios (ca. < 1 ppb) and was also associated with a westerly direction. At high wind speeds, the ND was also high between all instrument pairs (Fig. S6). As we observed high ND at relatively high wind speeds, it would suggest that spatial variability in ambient HONO concentrations did not affect the intercomparison as high wind speeds typically homogenize ambient concentrations from point and local sources. Overall from Figs. 3 and S5, the periods of low HONO concentration, high wind speeds and westerly winds all coincided during the formal winter intercomparison making it difficult to disentangle the influence of these factors on the observed ND.

There was evidence from the CV (Fig. 2) and ND (Fig. 3) analyses that the level of agreement between instruments decreased at low HONO mixing ratios. Therefore, we applied RMA correlation analysis for periods when the HONO level was below 2 ppb (as measured by CAM), and the results are shown in Table 4. From Table 4, the observed slopes between the BHAM–ICCAS–AIOFM at low concentrations (< 2 ppb) were similar to those for the whole winter intercomparison dataset (Table 2), unlike when compared to the CAM instrument. This suggests that the difference in measured concentrations between these instruments (BHAM–CAM–AIOFM, as indicated by the slope) was not related to concentration. The decrease in the slope for the low concentrations between CAM and the other three instruments compared to whole intercomparison (Tables 2 and 4, respectively), potentially points to changes in the CAM readings at lower concentrations. This change may be related to differences in instrument sensitivity (Table 1).

Measurements from the Manchester ToF-CIMS are compared to the BHAM and CAM instruments for the summer campaign as these instruments had the best data coverage for periods when the MANC instrument was measuring, as well as representing the typical upper and lower measurements (Fig. S9). In general the MANC instrument captured the temporal trends (r > 0.84) but recorded higher HONO concentrations than the other instruments (Table 7). Similar distributions were observed between the BHAM and MANC datasets, with the exception of a number of outliers for MANC (Fig. 4). We note that MANC was not co-located with either BHAM or CAM instrument and while this will likely have affected the intercomparison, the results do point to the MANC instrument capturing the temporal trends but at a higher concentration than the other instruments (157 %–239 %, Table 7).

The York SIFT-MS was primarily used for measuring VOC fluxes and so did not typically measure at ground level. To enable an intercomparison with the other techniques, the YORK instrument measured at ground level, from 18:00 30 May until 09:00 China standard time (CST). 31 May 2017. The results are shown in Fig. 5, and while the short time period and spatial distance between inlets (approx. 50 m) limits the conclusions that can be drawn, it is clear that the YORK instrument captured the temporal trends (r of 0.9–0.96 compared to other techniques) and gave comparable concentrations to the BHAM instrument (slope of 0.78). Furthermore, we note that a co-located PTR-MS (PTR-TOR 1000, Ionicon) was unable to see a HONO signal despite both instruments using H3O+ to detect HONO.