A Temporally and Spatially Resolved Validation of Emission Inventories by Measurements of Ambient Volatile Organic Compounds in Beijing, China

Understanding the sources of volatile organic compounds (VOCs) is essential for ground-level ozone and secondary organic aerosol (SOA) abatement measures. We made VOC measurements at 27 sites and online observations at an urban site in Beijing from July 2009 to January 2012. Based on these measurement data, we determined the spatial and temporal distribution of VOCs, estimated their annual emission strengths based on their emission ratios relative to carbon monoxide (CO), and quantified the relative contributions of various sources using the chemical mass balance (CMB) model. These results from ambient measurements were compared with existing emission inventories to evaluate the spatial distribution, species-specific emissions, and source structure of VOCs in Beijing. The measured VOC distributions revealed a hotspot in the southern suburban area of Beijing, whereas current emission inventories suggested that VOC emissions were concentrated in downtown areas. Compared with results derived from ambient measurements, the annual inventoried emissions of oxygenated VOC (OVOC) species and C2–C4 alkanes may be underestimated, while the emissions of styrene and 1,3-butadiene may be overestimated by current inventories. Source apportionment using the CMB model identified vehicular exhaust as the most important VOC source, with the relative contribution of 49 %, in good agreement with the 40–51 % estimated by emission inventories. The relative contribution of paint and solvent utilization obtained from the CMB model was 14 %, significantly lower than the value of 32 % reported by one existing inventory. Meanwhile, the relative contribution of liquefied petroleum gas (LPG) usage calculated using the CMB model was 6 %, whereas LPG usage contribution was not reported by current emission inventories. These results suggested that VOC emission strengths in southern suburban area of Bei-jing, annual emissions of C2–C4 alkanes, OVOCs and some alkenes, and the contributions of solvent and paint utilization and LPG usage in current inventories all require significant revisions.


VOCs sampling and analysis 2.1 Sampling sites for VOCs regional measurements in Beijing
Table S1 Summary of 27 sampling sites for regional VOC measurements in Beijing.
a The n is the number of regional sites classified in each site category.

VOCs analysis by online GC-MS/FID
Briefly, two parallel sampling channels collected two separate 300-mL aliquots of ambient samples cryogenically (at approximately -160°C), during the first 5 min of every hour. One channel was designed to measure C2-C5 hydrocarbons. Target

Sampling locations and periods for regional VOC measurements in Beijing
To minimize the impact of meteorological conditions on ambient trace gas levels, 16 sampling days, once monthly from July, 2009 to September, 2010 and January 2011, were selected carefully. Most sampling campaigns were conducted on heavy air pollution days characterized by low wind speed, high relative humidity, and poor visibility. However, the air quality on July 24 th , 2009 and February 4 th , 2010 was excellent, the former because a summer precipitation process had just ended in Beijing and the latter because of a recent cold winter northwestern airstream.
Sampling was conducted at 09:00, to assess atmospheric pollutant levels during morning traffic rush hour before the daily photochemical cycle, and 13:00, around the daily O 3 peak. This allowed temporal comparisons and an examination of the photochemical removal of reactive NMHC species and secondary production of ground-level O 3 and alkyl nitrates. Details of the sampling periods and meteorological parameters are summarized in Table S2.

VOCs analysis by offline GC-MS/FID
A 500-mL aliquot of air sample from each canister was concentrated using a three-stage cryofocusing pre-concentration system (Entech 7100, Entech Instruments) and analyzed with a GC (HP-7890A, Agilent Technologies, Santa Clara, CA, USA).
A PLOT (AL/KCL) column with a FID was used to separate and analyze most of the

Correlations between VOC species and CO
In order to check whether VOC emission ratios between day and night is different, correlations between several species with long atmospheric lifetime and CO at the PKU site during summer were analyzed. The calculated ratios for ethane, propane, n-butane, benzene, and acetone relative to CO using measurement data of 03:00-07:00 LT and all data agrees very well (Fig. S2), suggesting that VOC emission ratios during daytime and nighttime do not change.

Average emission ratios of VOC in Beijing
The average emission ratios for VOC at 27 sites in Beijing can be calculated by two approaches. The first method is to derive the slope of measured VOC mixing ratios to CO levels using the ODR linear regression fit for regional measurement data at 09:00 LT from all sites (the "linear regression" method). The standard deviations of linear regression slopes correspond to the standard deviations of average VOC emission ratios for regional sites. The relative standard deviations (RSD) of emission ratios for most VOC species were lower than 20%. The other method to calculate average value and standard deviation of VOC emission ratios in Beijing is based on the calculated VOC emission ratios at each regional site. For the VOC regional measurements in Beijing, there are two samples (09:00 and 13:00 LT) that were collected at each site during each month. VOC emission ratios at each site were calculated only based on 09:00 data due to the influence of photochemical processing on 13:00 data. Considering the limited data at each site (i.e. 6 data for winter, and 10 data for summer), the VOC emission ratio at each site was estimated as the average value of VOC/CO ratios in each sample (the "average" method). Average VOC emission ratios of Beijing calculated using "average" and "linear regression" methods showed a good agreement, with slope of 1.01 and r of 0.98 (Fig. S3). However, the relative standard deviations of VOC emission ratios calculated using the "average" method were larger than those determined by the "linear regression" method, with values in the range of 23-71%.