Detection of formaldehyde emissions from an industrial zone in the Yangtze River Delta region of China using a proton transfer reaction ion-drift chemical ionization mass spectrometer

A proton transfer reaction ion-drift chemical ionization mass spectrometer (PTR-ID-CIMS) equipped with a hydronium (H+3 O) ion source was developed and deployed near an industrial zone in the Yangtze River Delta (YRD) region of China in spring 2015 to investigate industryrelated emissions of volatile organic compounds (VOCs). Air pollutants including formaldehyde (HCHO), aromatics, and other trace gases (O3 and CO) were simultaneously measured. Humidity effects on the sensitivity of the PTR-IDCIMS for HCHO detection were investigated and quantified. The performances of the PTR-ID-CIMS were also validated by intercomparing with offline HCHO measurement technique using 2,4-dinitrophenylhydrazone (DNPH) cartridges and the results showed fairly good agreement (slope= 0.81, R2= 0.80). The PTR-ID-CIMS detection limit of HCHO (10 s, three-duty-cycle averages) was determined to be 0.9– 2.4 (RH= 1–81.5 %) parts per billion by volume (ppbv) based on 3 times the standard deviations of the background signals. During the field study, observed HCHO concentrations ranged between 1.8 and 12.8 ppbv with a campaign average of 4.1± 1.6 ppbv, which was comparable with previous HCHO observations in other similar locations of China. However, HCHO diurnal profiles showed few features of secondary formation. In addition, time series of both HCHO and aromatic VOCs indicated strong influence from local emissions. Using a multiple linear regression fit model, on average the observed HCHO can be attributed to secondary formation (13.8 %), background level (27.0 %), and industryrelated emissions, i.e., combustion sources (43.2 %) and chemical productions (16.0 %). Moreover, within the plumes the industry-related emissions can account for up to 69.2 % of the observed HCHO. This work has provided direct evidence of strong primary emissions of HCHO from industryrelated activities. These primary HCHO sources can potentially have a strong impact on local and regional air pollution formation in this area of China. Given the fact that the YRD is the largest economic zone in China and is dense with petrochemical industries, primary industrial HCHO emissions should be strictly monitored and regulated.

Abstract.A proton transfer reaction ion-drift chemical ionization mass spectrometer (PTR-ID-CIMS) equipped with a hydronium (H + 3 O) ion source was developed and deployed near an industrial zone in the Yangtze River Delta (YRD) region of China in spring 2015 to investigate industryrelated emissions of volatile organic compounds (VOCs).Air pollutants including formaldehyde (HCHO), aromatics, and other trace gases (O 3 and CO) were simultaneously measured.Humidity effects on the sensitivity of the PTR-ID-CIMS for HCHO detection were investigated and quantified.The performances of the PTR-ID-CIMS were also validated by intercomparing with offline HCHO measurement technique using 2,4-dinitrophenylhydrazone (DNPH) cartridges and the results showed fairly good agreement (slope = 0.81, R 2 = 0.80).The PTR-ID-CIMS detection limit of HCHO (10 s, three-duty-cycle averages) was determined to be 0.9-2.4(RH = 1-81.5%) parts per billion by volume (ppbv) based on 3 times the standard deviations of the background signals.During the field study, observed HCHO concentrations ranged between 1.8 and 12.8 ppbv with a campaign average of 4.1 ± 1.6 ppbv, which was comparable with previous HCHO observations in other similar locations of China.However, HCHO diurnal profiles showed few features of secondary formation.In addition, time series of both HCHO and aromatic VOCs indicated strong influence from local emissions.Using a multiple linear regression fit model, on average the observed HCHO can be attributed to secondary formation (13.8 %), background level (27.0 %), and industryrelated emissions, i.e., combustion sources (43.2 %) and chemical productions (16.0 %).Moreover, within the plumes the industry-related emissions can account for up to 69.2 % of the observed HCHO.This work has provided direct evidence of strong primary emissions of HCHO from industryrelated activities.These primary HCHO sources can potentially have a strong impact on local and regional air pollution formation in this area of China.Given the fact that the YRD is the largest economic zone in China and is dense with petrochemical industries, primary industrial HCHO emissions should be strictly monitored and regulated.

Introduction
Formaldehyde (HCHO) has been well recognized as one of the most abundant and important carbonyls in the atmosphere (Dasgupta et al., 2005;Lei et al., 2009;Li et al., 2014;Olaguer et al., 2013;Parrish et al., 2012;Rappenglück et al., 2010).HCHO is ubiquitous in the atmosphere with a typ-Published by Copernicus Publications on behalf of the European Geosciences Union.
Y. Ma et al.: Detection of formaldehyde emissions from an industrial zone ical near-ground concentration ranging from sub-ppbv levels in the rural area to ∼ 60 ppbv in highly polluted regions (Finlayson-Pitts and Pitts, 1999).HCHO plays a crucial role in atmospheric photochemistry.The absorption spectrum of HCHO extends well into the UVA region and thus its photolysis rate coefficient (Reaction R1) is one of the highest among all carbonyl compounds in the atmosphere (Gratien et al., 2007).Photolysis of HCHO (Reaction R1a) can contribute substantially to the atmospheric HO x radical (OH + HO 2 ) budget and thus can exert great impact on the oxidative capacity of the atmosphere (Volkamer et al., 2010).Since hydrogen (H 2 ) is classified as an indirect greenhouse gas, the R1b channel is also of atmospheric importance and may have a non-negligible impact on global climate change (Schultz et al., 2003;Tromp et al., 2003).
In the background atmosphere, HCHO is mostly formed from photochemical oxidation of methane (CH 4 ) and terminal alkenes (Finlayson-Pitts and Pitts, 1999).HCHO can also be emitted directly into the air from vegetation, although its formation mechanism is not well known (Seco et al., 2007).HCHO flux measurements conducted in forest areas indicate that both plants and the ground surface can contribute to assignable HCHO emissions, although the magnitude is relatively small compared to the photochemical production (DiGangi et al., 2011).In contrast to biogenic HCHO sources, anthropogenic activities are often intense primary HCHO sources.HCHO can be produced from any incomplete combustion processes.Vehicle exhaust is believed to be the dominant primary HCHO sources in heavily populated urban areas (Anderson et al., 1996;Ho et al., 2012).Regardless of the types of fuel, all internal combustion engines emit considerable amounts of HCHO.Interestingly, usage of alternative fuels can substantially reduce vehicle emissions of nitrogen oxides (NO x ) but may significantly increase emissions of HCHO and other carbonyl compounds (Wagner and Wyszynski, 1996).A recent study by Rappenglück et al. (2013) suggests that currently available mobile emission models (e.g., MOBILE6 and MOVES) may significantly underestimate the HCHO emissions from vehicles and there is a non-negligible difference of HCHO emissions between lightduty and heavy-duty diesel vehicles.Nevertheless, studies have found that vehicle exhaust alone may lead to significant underestimation of the emission inventory of carbonyl compounds in Beijing (Wang et al., 2014), most likely due to the complicated but poorly understood emissions from the industrial sector (Chen et al., 2014).High concentrations of HCHO have been observed around the Houston Ship Channel area populated with petrochemical facilities (Rappenglück et al., 2010).Using an imaging differential optical absorption spectrometer, Pikelnaya et al. (2013) have detected direct HCHO emission from burning flares of the refineries in Houston.In addition to flares, desulfurization and other petrochemical processes such as catalytic cracking of refinery feed may also contribute substantially to primary HCHO emissions (Olaguer et al., 2013).Other industry-related activities, such as wood processing and manufacturing of insect and fungus products, also contribute substantial primary HCHO emissions (Garcia et al., 2006).Recent studies have demonstrated that primary carbonyl emissions from a remote oil and gasproduction region can lead to high ozone (O 3 ) formation in a nearby rural area (Edwards et al., 2014).Therefore, it is of practical interest to fully comprehend the detailed primary emission sources of carbonyl compounds, especially in the case of HCHO.
Various source apportionment methods have been successfully utilized to identify primary and secondary HCHO sources (Chen et al., 2014;Garcia et al., 2006;Li et al., 2014;Olaguer et al., 2013;Parrish et al., 2012;Zheng et al., 2013b).Due to the lack of carbonyl emission inventories, particularly in China, it is impractical to directly estimate HCHO emissions using emission factors.However, carbon monoxide (CO), sulfur dioxide (SO 2 ), and toluene have been observed co-emitted with carbonyl compounds from industrial facilities at various locations and hence can serve as tracers of industrial HCHO emission (Pikelnaya et al., 2013;Rappenglück et al., 2010;Wang et al., 2015); i.e., variations of these gases are linearly associated with the primary HCHO emissions.Accordingly, correlation-based methods, such as the multiple linear regression fit model, can be used to quantify the relative contribution of primary industrial HCHO sources (Li et al., 2014).
HCHO can be measured by various analytical techniques (Gilpin et al., 1997;Li et al., 2005), among which proton transfer reaction mass spectrometry (PTR-MS) does not require sample pretreatments, is free from optical impairment due to heavy aerosol loading, and can potentially achieve a time resolution of a few seconds (Karl et al., 2003).It has been demonstrated that PTR-MS is particularly suitable for capturing strong point sources of volatile organic compounds (VOCs) (Zheng et al., 2013a, b).In this work, a proton transfer reaction ion-drift chemical ionization mass spectrometer (PTR-ID-CIMS) (Fortner et al., 2004;Zheng et al., 2008Zheng et al., , 2010Zheng et al., , 2015a) ) equipped with a corona discharge (CD) hydronium ion source (Zheng et al., 2015b) was developed to investigate VOC-related emissions from an industrial zone in Nanjing, China.The performance of the PTR-ID-CIMS was evaluated by intercomparing with the more established DNPH (2,4-dinitrophenylhydrazone) technique and the contribution of industry-related activities to primary HCHO emissions was evaluated with a multiple linear regression fit model.

Observation site
The measurements were carried out from 15 to 30 April 2015 on the campus of the Nanjing University of Information Science and Technology (NUIST).A detailed description of the site (32 • 12 20.8N, 118 • 42 19.2E) has been given in our previous work (please see Fig. 1 in Zheng et al. (2015a) for details).Briefly, the site was established by the Chinese Meteorological Administration as a training facility for meteorological observations.Meteorology parameters, including wind direction, wind speed, ambient temperature, pressure, relative humidity, and solar radiation, were continuously measured according to Chinese national standards (GB31221-2014).The site was located in the middle of two highways, i.e., G40 and G205, which were about 1.3 km to the west and 1.5 km to the east of the site, respectively.From the aromatics measurement results, we found no substantial impacts on measured HCHO from traffic-related emissions, which will be detailed in the discussion section.When easterly winds were dominant, the site was constantly affected by the outflow from the industry zone (Zheng et al., 2015a).All instruments were housed inside an air-conditioned trailer.The sample line was made of 6.4 mm OD PFA tubing and was installed on the roof of the trailer, about 6 m above ground level.

PTR-ID-CIMS
A schematic diagram of the PTR-ID-CIMS is shown in Fig. 1.The working principle of the PTR-ID-CIMS is fundamentally the same as a commercial Ionicon PTR-MS (Hansel et al., 1995).The PTR-ID-CIMS consists of a water reservoir, a point-to-surface CD ion source (Zheng et al., 2015b), a drift tube, and a quadrupole mass spectrometer (QMS) (Extrel, MAX 1000).During operation, 30.0 standard cm 3 min −1 (SCCM) of ultrapure nitrogen (N 2 ) regulated by a mass flow controller (Axetris AG) was fed into the water reservoir and then carried water vapor into the CD ion source, where a stainless steel needle biased with 1300 V discharged on the wall of a 2.5 cm long, 6.4 mm OD stainless steal tubing.Generated hydronium ions (H + 3 O) were pushed into the drift tube through a 1 mm diameter pinhole.The drift tube was comprised of 11 2.5 cm OD, 9.5 mm ID, 6.4 mm thick stainless steel rings that were separated from each other by 1.0 mm thick PTFE spacers.The metal rings were interconnected with 1.0 M resistors.On the sidewall of the first ring was mounted a PTFE critical orifice limiting the sampling flow to be 340 SCCM.The drift tube was pumped with an Agilent TS-300 dry scroll pump and the pressure in the drift tube was 2.0 mbar, measured by an Agilent CDG-500 capacitance diaphragm pressure gauge.The voltage between any two adjacent rings was 37.0 V, which produced an electric field of 52.9 V cm −1 .The corresponding ratio of electric field (E) to the buffer gas number density (N ), i.e., E / N ratio, was  108 Td (1 Td = 1 × 10 −17 V cm 2 ).To maintain a constant reaction temperature, the drift tube was regulated at 60 • C. A 200 µm pinhole (Edmund Optics) biased by 1 V was used to separate the drift tube from the high vacuum region that was housing the quadrupole mass analyzer and the Channeltron electron multiplier.The vacuum chamber was differentially pumped by two Agilent TV-301 turbomolecular pumps, which shared one Agilent IDP-3 dry scroll pump as the backing pump.Ambient air was delivered to the PTR-ID-CIMS through the 6 m long sample line and a diaphragm pump was used to pump the inlet at 10 standard L min −1 (SLPM) and thus to minimize the sample residence time no more than 1 s.The inlet was also heated to ∼ 60 • C to reduce potential wall losses.Using an automatic PTFE threeway valve, background checks were conducted once every 30 min for 10 min by rerouting sample air through a 350 • C Pt catalytic converter (Zheng et al., 2013b(Zheng et al., , 2015a)).
Figure 2 is a typical mass spectrum generated by the PTR-ID-CIMS after scanning the laboratory air by unit mass.Hydronium ion (m/z 19) and its water clusters (m/z 37, m/z 55, m/z 73, and m/z 91) can be observed clearly in the spectrum and are the dominant ions.Although the oxygen ion (m/z 32) is also present, it is less than 2.5 % of the primary hydronium ion (m/z 19) and thus does not interfere significantly with the PTR reactions initiated by hydronium ions.Masses 33, 42, and 59 can be respectively attributed to methanol, acetonitrile, and acetone because all of these chemicals are commonly utilized as organic solvents in the organic chemistry laboratories in our department.We want to point out that mass 33 can also originate from isotopic peak of O + 2 with oxygen-17 ( 17 OO + ) and protonated oxygen (HO + 2 ), which, however, are also present during background checks using a catalytic converter and thus both of them will be subtracted from the methanol signal.Mass 59 can also originate from propanal and glyoxal.However, since both propanal and especially glyoxal are significantly more reactive than acetone in the atmosphere, both of them typically contribute less than 10 % of the mass 59 signal (de Gouw and Warneke, 2007).The other peaks were most likely due to the chemicals and their fragments originated from various chemistry experiments in the building.
In theory, the PTR-ID-CIMS worked essentially in the same way as an Ionicon PTR-MS in VOC detections.However, the PTR-ID-CIMS was developed with an intention to do more than positive ion chemical ionization analysis and thus it was equipped with a bipolar ion detection system.It can be readily converted into a negative ion CIMS.With the proper ion chemistry scheme, the PTR-ID-CIMS can be used to detect not only VOCs but also ammonia (NH 3 ) (Nowak et al., 2006), nitrous acid (HONO) (Pinto et al., 2014), and nitric acid (HNO 3 ) (Zheng et al., 2008).In the future, the drift tube can also be modified to adapt an atmospheric pressure interface (API) to do sulfuric acid measurements (Zheng et al., 2010).Another significant difference between the PTR-ID-CIMS and the Ionicon PTR-MS was the inlet system.In the PTR-ID-CIMS air samples were introduced into the drift tube through a critical orifice instead of a long (> 1 m) 1.5 mm OD capillary PEEK tubing used in the Ionicon PTR-MS.We found our inlet setup had the advantages of faster time response and lower wall losses and especially suited for sticky gas measurements, such as NH 3 (Zheng et al., 2015a).In this work, we focus on the measurement and analysis of VOC emissions in the Yangtze River Delta (YRD) region using the PTR-ID-CIMS, especially from industry-related sources.

HCHO Measurement with PTR-ID-CIMS
The PTR-ID-CIMS was operated in the single ion monitor (SIM) mode; i.e., a series of masses were sequentially detected by the PTR-ID-CIMS.The QMS was set to measure each mass for 2 s and then pause for 2 s after switching to a new setting to account for the slew time of the QMS power supply.More than 40 masses were measured during each measurement cycle and it took about 3 min to finish one duty cycle.HCHO was detected at m/z 31.Because the proton affinity of HCHO is slightly higher than water, Reaction (R2) is reversible.
Therefore, the HCHO measurement using PTR-ID-CIMS was strongly affected by the water concentration in the drift tube.This water dependency has been investigated in details previously and the sensitivity dependence on water in HCHO detection can be evaluated as the following (Inomata et al., 2008;Jobson and McCoskey, 2010;Schripp et al., 2010;Warneke et al., 2011;Wisthaler et al., 2008;Zheng et al., 2013b): where k and k are the forward and reverse reaction rate coefficients of Reaction (R2), respectively, and t is the ionmolecule reaction time.Given Eq. ( 1) can be simplified and rearranged as Eq. ( 2): The left-hand side of the equation indicates that H + 3 Onormalized H + HCHO counts rate signal (normalized counts s −1 , NCPS) with respect to 1 ppbv of HCHO (NCPS ppbv −1 ) or that the sensitivity of the PTR-ID-CIMS to HCHO is a function of water vapor concentration inside the drift tube in the form of Warneke et al. (2011): where x is the water vapor concentration, A = k/k , and B = k t.Therefore, the sensitivity of the PTR-ID-CIMS to HCHO at certain humidity can be evaluated according to Eq. ( 2) if the parameters A and B are known.Figure 3 shows a plot of the PTR-ID-CIMS response to HCHO standards under various ambient humidities.Each data point of Fig. 3 was the result of one set of calibration, which will be described in details in the following section.The corresponding water vapor concentration was determined from sample air humidity and the ion source carrier flow humidity.The values of A and B were inferred from curve fitting using Eq.(3) as the fitting function.The linear correlation coefficient between the original and fitted values was determined to be 0.82.In addition, given that the reaction temperature and E / N in this work were kept constant (i.e., 60 • C and 108 Td) and the reduced ion mobility of H + 3 O in air was taken as 2.76 cm 2 V −1 s −1 (de Gouw et al., 1997), the ion-molecule reaction time was determined to be about 103 µs.Accordingly, k and k of Reaction (R2) were also evaluated from A and B with values of 0.84 × 10 −9 cm 3 molecule −1 s −1 and 2.0 × 10 −11 cm 3 molecule −1 s −1 , respectively.The energy dependency of k and k has been investigated by Hansel et al. (1997).It was demonstrated that the reverse reaction channel, k , increased significantly with increasing mean relative kinetic energy (KE) of the reactants while k only showed slightly negative energy dependency.KE under the reaction conditions of this work was calculated according to the method as described by Hansel et al. (1997).The reduced ion mobility of H + HCHO in air was assumed to be similar as that of H + 3 O q H 2 O (2.27 cm 2 V −1 s −1 ) (de Gouw et al., 1997), since they have a similar molecular size.Therefore, the KEs associated with the forward and the reverse reactions were respectively determined to be 0.12 eV and 0.10 eV.The corresponding k and k measured by Hansel et al. (1997) were 1.4 × 10 −9 and 1.1 × 10 −11 cm 3 molecule −1 s −1 , respectively.Evidently, the inferred values from Fig. 3 agreed fairly well with the literature values.x (1 − e −Bx ), deduced from Eq. ( 1), where A is the ratio between the forward (k) and reverse (k ) reaction rate coefficient and B is the product of k and the ion-molecule reaction time t.
In Fig. 3 much less variation in instrument sensitivity was observed in this work than the previous studies (Vlasenko et al., 2010).The most likely reason is that the ion source of the PTR-ID-CIMS used humidified pure nitrogen as carrier gas.The nitrogen flow rate (30.0 SCCM) was much higher than the water vapor flow rate used by a typical Ionicon PTR-MS and more importantly all of the humidified nitrogen was sucked into the drift tube instead of being pumped away in the case of PTR-MS.Consequently, the background water concentration in the drift tube of the PTR-ID-CIMS was significantly higher than a typical PTR-MS.As indicated by Fig. 5 of Vlasenko et al. (2010), the instrument sensitivity will change substantially when the water content changes slightly from the completely dry condition.Therefore, the relatively high background water content in the PTR-ID-CIMS can lead to decreased sensitivity during "dry" calibrations, which can explain the much lower variation in sensitivity observed in this work.

PTR-ID-CIMS calibration
PTR-ID-CIMS calibrations were performed by mixing gaseous VOC standards (Apel-Reimer Environmental, USA) into zero air through the calibration port.The VOC standard flow rate was controlled by a metal-seal 100 SCCM mass flow controller (UNIT, UFC-1260A).The mixing ratios of the target VOCs inside the cylinder are on the order of a few hundreds of ppbv with an uncertainty of < 5 %; e.g., the mixing ratios of benzene and toluene were 208 and 157 ppbv, respectively.Before each calibration, the PFA tubing to deliver the VOC standard gas was passivated by flowing a few SCCM of VOC standards through it for at least 8 h.The zero air was normally generated in situ by passing ambient air www.atmos-meas-tech.net/9/6101/2016/Atmos.Meas.Tech., 9, 6101-6116, 2016 through the catalytic converter and its flow rate was controlled by a critical orifice (as shown in Fig. 1), which was precisely quantified by a Gillibrator (Sensidyne Gilian).Ultra pure nitrogen was also used in place of the zero air to conduct calibrations under the dry condition.The HCHO standard concentration in the cylinder was verified by the DNPH cartridge measurements with a value of 284 ± 2 ppbv (based on three independent measurements).Background signals were checked before and after the HCHO standard additions using the catalytic converter.The PTR-ID-CIMS was also operated in the SIM mode during calibrations.Similar to the field measurement cycle, the integration time at each mass was set to 2 s and the time interval between two consecutive masses was also set to 2 s.Each calibration cycle contained 48 masses, including m/z 21 (H 18 3 O + ), m/z 25 (system background), m/z 30 (NO + ), m/z 32 (O + 2 ), m/z 37 (H 3 O + H 2 O), and other VOC standard gases.It took about 3 min to complete one measurement cycle.Figure 4a shows the time series of a typical HCHO calibration.Since the nitrogen-15 isotopic peak of NO ( 15 NO + ) also contributes to the signal at m/z 31, relatively high background signals were observed in Fig. 4a.In addition, because the quadrupole mass filter can normally achieve unit mass separation, m/z 31 signals were also slightly interfered by the tail from the relatively strong adjacent oxygen peak (O + 2 ), which was about 1 to 2 % of the H + 3 O intensity.Although 15 NO + interference can be removed from the real HCHO signal by background subtraction, to accurately reflect HCHO background signal variations, 0.37 % (Hoffmann and Stroobant, 2007) of the signal of m/z 30 (not shown in Fig. 4a) was subtracted from HCHO background check signals before the background noise was evaluated.Each insert in Fig. 4 is the corresponding calibration curve with background signal subtracted.To fully characterize the humidity effects on its performance, the PTR-ID-CIMS was also calibrated under various relative humidity conditions (1.0, 17.5, 29.0, 47.0, 62.5, and 81.5 %) using pure air as the carrier gas.The relative humidity of the VOC standards was achieved by passing the carrier gas through a water bubbler similar to the one used in Zheng et al. (2015a).Three to four consecutive calibrations were performed at each RH setting.The results are displayed in Fig. 5.It is clearly shown that higher instrument sensitivity (i.e., NCPS per ppbv HCHO) is associated with lower RH setting.The instrument sensitivity and the corresponding detection limit (DL) (based on 3 times the standard deviations of the background signals) at each RH setting were evaluated from each individual calibration.Furthermore, the instrument precision at each RH setting was assessed by the variations of both the instrument sensitivity and the DL among repeated calibrations.The results showed that on average the PTR-ID-CIMS DL of HCHO varied from 0.9 ppbv for dry condition to 2.4 ppbv for 81.5 % RH at room temperature.Among all RH settings (< 81.5 %) the variations of the instrument sensitivity and the DL were less than 17 and 16 % of the average values, respectively.Therefore, for the worst case, the HCHO measurement uncertainty in this work should be < 18 %, including the 1 % uncertainty associated with the HCHO standard concentrations.The performances of various HCHO measurement techniques including this work have been listed in Table 1.Evidently, the sensitivity of PTR-ID-CIMS for HCHO detection was comparable to other PTR-MS-based techniques and was fast enough to capture pollution episodes with a timescale of at least a few hours.
Calibrations of benzene (Fig. 4b) and toluene (Fig. 4c) were conducted using the VOC standard mixture simultaneously with the HCHO calibrations.For three-duty-cycle averages, the calibration factors and detection limits were, respectively, 41.2 NCPS ppbv −1 and 0.06 ppbv for benzene and 40.0 NCPS ppbv −1 and 0.07 ppbv for toluene.In a similarly way as the case of HCHO, we determined the uncertainties of benzene and toluene measurements to be less than 8.7 and 11.0 % (including the 5 % uncertainty associated with the VOC standard concentration) in this work.

DNPH HCHO measurements
Carbonyls including HCHO were also measured with the DNPH method followed the US EPA method TO-11A.Cartridges coated with DNPH (Agela Technologies, Tianjin) were used to scrub carbonyls from the ambient air samples and then analyzed by high-performance liquid chromatography (HPLC) (Waters, Alliance e2695) equipped with a Diamonsil C-18 column (5 µm, 250 mm by 4.6 mm).The cartridge was attached to the front end of a 6.4 mm OD PFA tubing that was bound to the inlet of the PTR-ID-CIMS.A potassium iodide (KI) cartridge was also installed in the front of the DNPH cartridge to remove ozone.Ambient air was pulled into the cartridges by a diaphragm pump and the flow rate was set to 0.53 SLPM by a critical orifice, which was calibrated with the Gillibrator.The sampling time varied from 0.6 to 2 h during the campaign depending on the ambient HCHO concentrations.The sampled cartridges were stored in Teflon bags at 4 • C right after sampling.Note that a few high O 3 episodes were encountered during the measurement period with a maximum concentration of about 140 ppbv, which can potentially interfere with DNPH samples.However, as demonstrated by the US EPA, the KI cartridges can efficiently remove 125-200 ppbv O 3 from air samples for up to 100 000 ppbv h with a requirement of minimum moisture level of 10 % RH (https://www3.epa.gov/ttnamti1/files/ambient/airtox/to-11ar.pdf, accessed October 2016).Given the relative high RH conditions during the campaign period, O 3 should not interfere with our DNPH HCHO measurements.
All cartridge samples were processed within 2 weeks from dates samples were collected.Each sample was first slowly eluted with 5 mL acetonitrile (HPLC grade) and 20.0 µL of the extract solution was injected into the HPLC Table 2. Parameters of the multiple linear regression fit and the linear correlation coefficient (R 2 ) for the measured and source-apportioned HCHO in Eq. ( 4).β 1 denotes the portion of HCHO from photochemical production.β 2 , β 3 , and β 4 are the emission ratios of HCHO with respect to CO, benzene, and toluene, respectively.β 5 represents HCHO concentration in the background atmosphere.

Other trace gas measurements
CO (Thermo Scientific, model 48i) and O 3 (Thermo Scientific, model 49i) were also measured at the site.Their operation and calibration procedures followed the manufacturer's instructions and have been detailed in previous work (Zheng et al., 2015a).
3 Results and discussion

Overall observation results
Meteorological parameters including wind direction and wind speed, relative humidity (RH) and ambient temperature (T ), and solar radiation are shown in Fig. 6a-c.During the observation period, the weather was mostly clear with occasional precipitation events on 18, 19, 27, and 28 April.RH and T showed typical anticorrelation in their diurnal variations.No gusty wind was experienced during the observation period and the average wind speed was ∼ 2.1 m s −1 .The time series of O 3 , CO, benzene (C 6 H 6 ), toluene (C 7 H 8 ), and HCHO (three-duty-cycle averages) are shown in Fig. 6d-f, respectively.Also shown in Fig. 6f (black dots) are the HCHO results from the DNPH cartridge measurements.The gaps in the trace gas measurements were due to either power interruptions or instrument calibrations.High HCHO concentrations up to 12.8 ppbv (three-duty-cycle averages) were observed in a few pollution episodes.All observed gases showed no obvious diurnal variations, except O 3 , which is a secondary air pollutant mainly formed from photochemical processes (Finlayson-Pitts and Pitts, 1999).Although HCHO was often considered as an intermediate oxidation product of VOCs, such as olefins and CH 4 , we did not observe noticeable diurnal pattern of HCHO in this work either.However, positive correlations (R 2 = 0.2-0.4) were found among HCHO, aromatics, and CO and significantly better correlations (R 2 = 0.4-0.6)among these species could  be found within individual plumes, which strongly suggests that primary emissions were responsible for the observed high concentrations of HCHO.In addition, all heavy pollution episodes observed in this campaign were associated with easterly wind, which can be demonstrated by the rose plots of HCHO, CO, benzene, toluene, and O 3 in Fig. 7.
The temporal variations of HCHO in this work also differed distinctly from that observed at the Tijuana site during the Cal-Mex 2010 campaign (Zheng et al., 2013b).Although both sites were located in the vicinity of industrial zones, the type of industry in Tijuana was dominated by electronic industries, which normally utilize large quantities of organic solvents (such as methanol, acetone, and ethyl acetate) for electronic parts cleaning during lithography processing (Zheng et al., 2013a).Assisted by the stronger subtropical solar radiation in May/June 2010 during Cal-Mex than in April 2015 during this study, stronger solar radiation can result in more HCHO formation from VOC oxidation in Tijuana.In contrast, the industrial zone in Nanjing is a rather intensive point source on a regional scale as indicated by Fig. 1 of Zheng et al. (2015a).It normally takes several hours for the plumes from the industrial zone to arrive at the site and thus it is reasonable to assume that the plumes are well mixed.The VOC pollution episodes observed in this work typically lasted for more than a few hours, which was consistent with previous ammonia and amine measurements conducted at the same site (Zheng et al., 2015a).The atmospheric lifetime of HCHO and aromatics are also on the order of at least a few hours.Therefore, we are confident that the 3 min time resolution should be sufficient to capture the variabilities of VOC plumes in this study.

A case study
To illustrate the performance of the PTR-ID-CIMS in more details and to establish a clear connection between VOC plumes and primary industrial emissions, we chose the measurements on 18 April for a case study.On 18 April, it was mostly cloudy with scattered light rain.Therefore, secondary formation of HCHO was substantially suppressed.Figure 8 displays the wind direction and wind speed (Fig. 8a), m/z 37/19 and RH (Fig. 8b), O 3 and CO (Fig. 8c), m/z 79 and m/z 93 in NCPS (Fig. 8d), and m/z 31 in NCPS (Fig. 8e).It is evident that all air pollutants shown in Fig. 8 were substantially elevated in the plume but exhibit different variation profiles with periods significantly longer than the measurement duty cycle (about 3 min).These fine structures in pollutant time-series indicate that these species were originated from the same area but might be associated with different industrial activities.
It is evident that humidity inside the PTR-ID-CIMS did not change significantly during background checks.Figure 8d and 8e are the time series of m/z 31, m/z 79, and m/z 93 raw data.The background signals of m/z 31 appear to be independent of ambient RH.We also want to point out that there is a significant difference between this work and that of Warneke et al. (2011).Warneke et al. (2011) conducted an airborne measurement of HCHO.Their PTR-MS would experience rapid humidity and pressure changes when the flight altitude changed from ground level to the free troposphere as shown in their Fig.3c.The diurnal variation of m/z 37/19 in this work was about 0.2 to 0.24 and no sig-nificant background changes were observed, which is consistent with the results of Fig. 3c in Warneke et al. (2011).From 10:00 LT on 18 April, there was a persistent easterly wind with wind speed varying within a few meters per second.Meanwhile, significant increase of HCHO was observed at the site.In particular, after 19:00 LT HCHO apparently anticorrelated with wind speed.At about 23:00 LT wind direction suddenly switched to westerly wind and wind speed increased to 3-5 m s −1 , bringing in cleaner air masses to the site.HCHO suddenly decreased to 2 ppbv.This observation was also consistent with the rose plots in Fig. 7. Evidently, HCHO observed on the night of 18 April can only be explained by primary sources other than automobile emissions.

Intercomparison with DNPH cartridge measurements
The PTR-ID-CIMS HCHO measurements were intercompared with and verified by the well-established DNPH method.Since the sampling time of the DNPH cartridges was much longer than 10 min, the CIMS data were averaged based on the time stamp for the DNPH cartridge measurements for the intercomparisons.Overall, both measurements agreed with each other fairly well (Fig. 6f). Figure 9 shows a scatter plot of PTR-ID-CIMS versus DNPH with a slope of 0.81 and an intercept of 0.66 (R 2 = 0.80, based on orthogonal distance regression).Based on the laboratory calibrations conducted with both N 2 and air (at various levels of RH), we found that the uncertainty of PTR-ID-CIMS HCHO measurements was within 18 % and that of he DNPH measurewww.atmos-meas-tech.net/9/6101/2016/Atmos.Meas.Tech., 9, 6101-6116, 2016 ments was within 3.6 %.Therefore, it is reasonable to believe that the observed discrepancy between these two data sets can be explained by the combined measurement uncertainties and the higher background level in the PTR-ID-CIMS measurements.

Determination of industrial emissions
As the observation site is located less than 10 km away from the east of an industrial zone, the site was constantly experiencing plumes originated from various industrial activities, such as crude oil refining, plastic and rubber syntheses, and pesticide productions.Since benzene, toluene, and other aromatics are heavily used in many chemical syntheses, benzene and toluene are mass produced in the industrial zone by both catalytic cracking of petroleum fractions and coal carbonization.It is clearly shown in Fig. 8d that toluene and benzene peaks often did not concur simultaneously, indicating that they were emitted from different facilities.This is also consistent with the relatively weak correlations (R 2 = 0.2-0.4)among individual species observed within the measurement period.Flue gases generated at these facilities were usually incinerated before they were discharged into the atmosphere.In fact many flares could be visually identified within the industrial zone.Both HCHO and CO were produced during incomplete combustion processes and direct evidence of HCHO emissions from flares have been confirmed by previous observations at oil refineries in Houston, Texas (Pikelnaya et al., 2013).Although it is well known that HCHO and CO were present in vehicle exhaust, the benzene to toluene ratio within CO plumes was determined to be 1.7 ± 1.0 and significantly higher than 0.5, an indicator of automobile emissions (Hoque et al., 2008), suggesting that these air masses were not dominated by vehicle exhaust and thus most of the benzene and toluene observed in this work were not likely co-emitted by vehicle exhausts.Therefore, HCHO, CO, benzene, and toluene were likely co-emitted from the industrial zone, although they may be originated from different industrial processes and transportrelated activities, such as the heavy-duty trucks operating in the vicinity of the industrial zone.Thus both benzene and toluene were selected as tracers of petrochemical-industryrelated sources and CO was selected as the tracer of any combustion processes including flares and vehicle exhaust.Since O 3 was from secondary formation, O 3 was used to represent secondary HCHO.Accordingly, the observed HCHO was ap-portioned by a multiple linear regression fit model (Eq.4): where C HCHO , C CO , C Benzene , and C Toluene are measured mixing ratios of HCHO, O 3 , CO, benzene, and toluene, respectively.β 1 , β 2 , β 3 , β 4 , and β 5 are coefficients obtained from the multiple linear regression fit.β 1 denotes the portion of HCHO from photochemical production.β 2 , β 3 , and β 4 are the emission ratios of HCHO with respect to CO, benzene, and toluene, respectively.β 5 represents HCHO concentration in the background atmosphere.According to previous work conducted at a rural site in the YRD region (Wang et al., 2015), the background level of HCHO was constrained to 1.00 ppbv to represent the regional conditions.The multiple linear regression fit analysis was performed with a userdefined fitting function (in the Igor Pro (version 6.36) software package with the same form as Eq. 4).The data used for analysis were the same three-duty-cycle-averaged data set as that presented in Fig. 6.The fitting process was initiated by assigning all five parameters (β 1 -β 5 ) with an initial value of 0.1.After the first run the fitting process was reiterated with the fitting results obtained from the previous run, except that β 5 was constrained to 1 ppbv by default to reflect the regional background conditions.After three iterations, the fitting results did not change from the previous run significantly and the associated standard deviations of the fitted parameters were very close to or less than 0.1.At that point, we considered the fitting results to be acceptable.To verify the usage of 1.00 ppbv as the background HCHO level, we also have performed a series of multiple linear regressions with constrained background levels ranging from 0.5 to 1.7 ppbv.The corresponding R 2 ranged from 0.43 to 0.53.In all cases, the parameters of β 3 and β 4 in Eq. ( 4) did not change sub-  stantially and remained the dominant terms.Since 1.00 ppbv HCHO background level was supported by previous work (Wang et al., 2015), we thus preferred 1.00 ppbv as the constrained background level.The fitting results are listed in Table 2.The sourceapportioned and measured HCHO time series are shown in Fig. 10.In general, the approximated results can cap-ture the trends of measured HCHO profiles fairly well, especially during the pollution episodes.A linear regression fit (Fig. 11) between approximated and measured HCHO shows reasonable agreement with a slope of 0.88 and R 2 of 0.52.Based on a Student t test, the measured and approximated HCHO values are not significantly different from each other (p value = 0.15) at a 0.05 significance level.The relative contributions of different sources to ambient HCHO are also tabulated in Table 3.Throughout the campaign period, secondary formation accounted for the smallest portion of the observed HCHO regardless within or out of the pollution episodes.The contribution associated with background level of HCHO accounted for about 22 to 29 % of the total HCHO.CO, benzene, and toluene represented various industry-related activities, such as oil refineries, petrochemical syntheses, power generations, flares, and transportation, which overall contributed substantially (59.2 %) to the total HCHO budget; this portion reaches 69.2 % when only pollution episodes are considered.The possible reason was that the observation site in this work was in the vicinity of the primary emission sources, which constantly emitted a large quantity of HCHO into the air.These sources were so strong that they can dominate the local photochemical production of HCHO and suppress the typical diurnal variations of HCHO.In theory, CO and aromatics also represented industry-related vehicle emissions in this work, which may account for a significant portion of the vehicle fleet emission in the studied area.However, to specify the portion of vehicles emission cannot be achieved by the current available tracers.The positive correlations between HCHO and these aforementioned air pollutants established a statistical link between HCHO and industrial activities.The relatively weak correlation of R 2 = 0.52 was most likely due to the complexity of the petro- chemical industry.Many organic synthesis processes may also lead to HCHO emissions.These tracers used in this work may not represent all processes that contributed to HCHO emissions.More studies are needed to further investigate the potential primary sources of HCHO in the industrial zone and to fully characterize industrial VOCs emissions.

Conclusions
High concentrations of HCHO were observed using a custom-built PTR-ID-CIMS at a suburban site in Nanjing, China, near an industrial zone, typical in the YRD region.The humidity dependency of the PTR-ID-CIMS sensitivity to HCHO was evaluated by systematic calibrations.The PTR-ID-CIMS measurement of HCHO was verified by the US EPA recommended offline DNPH method (TO-11A).Within the observation period, HCHO ranged between 1.8 and 12.8 ppbv with a campaign average of 4.1 ± 1.6 ppbv, which was comparable with previous HCHO observations in other similar locations.Temporal variations of HCHO showed no clear correlation with O 3 , indicating the presence of strong primary point sources of HCHO.All high HCHO episodes observed in this study were associated with air masses originated from the east of the observation site where the industry zone is located and populated with heavy industrial facilities especially petrochemical-related manufacturers.Furthermore, other primary air pollutants including CO, benzene, and toluene were simultaneously measured within the plumes and showed significant correlations with HCHO.Benzene, toluene, and other aromatics were commonly used and mass produced in petrochemical industries (such as catalytic cracking of petroleum fractions and coal carbonization) and hence were excellent tracers of petrochemical processes.Both CO and HCHO can be produced during incomplete combustions, such as flares and vehicle exhaust, and thus CO can be considered a tracer of any combustion processes.However, the benzene to toluene ratio was found significantly higher than 0.5 in this work, indicating that vehicle exhaust may not be a major contributor to HCHO.Using O 3 , CO, benzene, and toluene as tracers, a multiple linear regres-sion analysis revealed that secondary formation, industryrelated activities (flares, various petrochemical productions, and transportation), and the background atmosphere respectively contributed 13.8, 59.2, and 27.0 % to the observed HCHO.Moreover, within the plumes the portion of industryrelated emissions can account for up to 69.2 % of the observed HCHO.This work has provided direct evidence of strong primary emissions of HCHO from industry-related activities in China.These primary HCHO sources can potentially have strong impact on local and regional air pollution formation.Given the fact that the YRD region is the largest economic zone in China and is dense with petrochemical industries, primary industrial HCHO emissions should be fully investigated in this area.HCHO emission ratios associated with individual sources are especially needed.

Data availability
The PTR-ID-CIMS data used in this work are available from the authors upon request (zheng.jun@nuist.edu.cn).

Figure 2 .
Figure 2. A typical mass spectrum produced by the PTR-ID-CIMS in the laboratory.

Figure 3 .
Figure3.The response of the PTR-ID-CIMS to HCHO standards (in the units of counts s −1 ppbv −1 HCHO standard per million H + 3 O ion, NCPS ppbv −1 ) as a function of water vapor concentration inside the drift tube.The correlation is in the form of y = Ax (1 − e −Bx ), deduced from Eq. (1), where A is the ratio between the forward (k) and reverse (k ) reaction rate coefficient and B is the product of k and the ion-molecule reaction time t.

Figure 4 .
Figure 4. Time series of a typical set of calibration of HCHO (a), benzene (b), and toluene (c).The insert in each panel is the calibration curve obtained from the corresponding calibration data.

Figure 5 .
Figure 5. HCHO calibrations conducted at RH = 1-81.5%.Three or four repeated calibrations were performed at each RH setting.The averaged background signals have been subtracted from all calibration data sets.The linear fitting lines are based on the average of each group of calibrations at one RH setting and the corresponding error bars represent 1 standard deviation of each group of calibrations.

Figure 7 .
Figure 7. Rose plots of HCHO, CO, benzene, toluene, and O 3 .The radius represents the percentage of each air pollutant within a certain mixing ratio range.The spread angle denotes the wind direction.

Figure 8 .Figure 9 .
Figure 8.Time series of wind direction (WD) and wind speed (WS) (a), m/z 37/19 ratio and RH (b), O 3 and CO (c), m/z 79 and m/z 93 (d), and m/z 31 signal (e) on 18 April 2015.The dips in (d) and (e) are the periods when background checks were made.

Figure 10 .Figure 11 .
Figure 10.Comparisons of the measured HCHO and source-apportioned HCHO from the multiple linear regression fit.Measured HCHO is denoted by black dots and contributions from different sources are shown as color-coded bars.

Table 1 .
Performance intercomparison of various HCHO measurement techniques.Based on the uncertainty of the instrument sensitivity.c Averaged over three duty cycles.

Table 3 .
Relative contributions (%) of different sources to the observed HCHO calculated using the multiple linear regression model.Pollution episodes are defined as periods when HCHO concentration was higher than the campaign average, i.e., 4.1 ppbv.