Whole-air samples were collected in 2 L electro-polished stainless-steel canisters, which had been cleaned and vacuumed according to the TO-15 method issued by the US EPA before shipment to the sampling sites. The restricted grab sampler (39-RS-x; Entech), which has a 5 µm Silonite-coated metal particulate filter, was placed on the inlet of the canister to completely filter out dust and other particulates during sampling. These samples were taken in daytime from 08:00 to 19:00 LT (Beijing time, GMT + 8) with an interval of 1 to 2 h. The sampling time was 5 min to fill the vacuumed canisters. The filled canisters were transported back to the laboratory of the Guangzhou Institute of Geochemistry, Chinese Academy of Science. Each air sample was analyzed for 65 light non-methane hydrocarbons (NMHCs) species. The samples were pretreated by an Entech Model 7100 Preconcentrator (Entech Instruments Inc., California, USA) and analyzed by a gas chromatography–mass selective detector (GC-MSD/FID, Agilent 7890A/5973N, USA) using dual columns and dual detectors to simultaneously analyze both low- and high-boiling-point volatile organic compounds (VOCs) with each injection. The detailed analytical procedure is described by Zhang et al. (2012).

In this study, we analyze 65 measurements of aromatics (benzene, toluene, ethyl-benzene, and m/p/o-xylene) and wind speed measurements taken across Tibet during October 2010 (Fig. 1a). Care was taken in sampling such that there are no direct urban, industrial, or road emissions in the upwind direction of the sampling location. The lifetimes of toluene, ethyl-benzene, and m/p/o-xylene are relatively short (2–20 h), and these reactive aromatic compounds therefore provide observational constraints for transport from India and nearby regions to Tibet. We group the samples into three periods based on the time and locations of the measurements, i.e., period 1 from 13 to 17 October 2010 to the north of the Himalayas along the southern border of Tibet, period 2 from 19 to 24 October across the interior of the Tibetan Plateau, and period 3 of 25 October to the west of the Yarlung Tsangpo Grand Canyon in southeastern Tibet (Fig. 1a).

The SCanning Imaging Absorption spectroMeter for Atmospheric CHartographY (SCIAMACHY) onboard Environmental Satellite (ENVISAT) operated from 2002 to 2012 (Burrows et al., 1995; Bovensmann et al., 1999), with an overpass time at about 10:00 local time. SCIAMACHY made passive remote-sensing measurements of the upwelling radiation from the top of the atmosphere in alternate nadir and limb-viewing geometry. Mathematical inversion of the measurements of SCIAMACHY yields a variety of data products including glyoxal (CHOCHO) vertical column densities (VCDs; unit: molecules cm-2). The retrieval uses the differential optical absorption spectroscopy (DOAS) technique (Wittrock et al., 2006; Vrekoussis et al., 2009; Alvarado et al., 2014). The CHOCHO retrieval used in this study is based on the algorithm developed in Alvarado et al. (2014), which includes corrections for the interferences with nitrogen dioxide (NO2) over the regions with high NOx emissions as well as liquid water over oceans (Alvarado, 2016). The detection limit for SCIAMACHY CHOCHO VCD is about 1×1014 molecules cm-2. The overall monthly uncertainty of CHOCHO VCDs (CCHOCHOSCIAMACHY) in the selected region during October 2010 is given by α×CCHOCHOSCIAMACHY+1×1014 molecules cm-2, where the value of α is in a range of 0.1 to 0.3. Following the method as described by Liu et al. (2012b), we derive a top-down aromatic emission estimate for South Asia constrained by CHOCHO retrievals described in Sect. 2.4.

We use the 3-D Regional chEmical trAnsport Model (REAM) to examine the chemistry evolution and regional transport of aromatics. REAM has been used in previous studies, including large-scale transport (Wang et al., 2006; Zhao et al., 2009b, 2010), vertical transport (Zhao et al., 2009a; Zhang et al., 2014, 2016), emission estimates (Zhao and Wang, 2009; Liu et al., 2012b; Gu et al., 2013, 2014, 2016), and other air quality studies (Zeng et al., 2003, 2006; Choi et al., 2005, 2008a, b; Wang et al., 2007; Liu et al., 2010, 2012a, 2014; Gray et al., 2011; Yang et al., 2011; Zhang and Wang, 2016).

REAM has a horizontal resolution of 36 km with 30 vertical levels in the troposphere and 5 vertical levels in the stratosphere covering adjacent regions of China (Fig. 1b). The model top is at 10 hpa. Meteorological fields in REAM are obtained from the Weather Research and Forecasting model (WRF) assimilations constrained by National Centers for Environmental Prediction Climate Forecast System Reanalysis (NCEP CFSR; Saha et al., 2010) 6-hourly products, which have a horizontal resolution of T382 (∼ 38 km). We run the WRF model with the same resolution as in REAM with a domain larger than that of REAM by 10 grid cells on each side. Meteorological inputs related to convective transport are updated every 5 min, while the others are updated every 30 min. The recent update of REAM expands the GEOS-Chem standard chemical mechanism (V9-02) to include a detailed description of aromatic chemistry (Bey et al., 2001; Liu et al., 2010, 2012b). Aromatics are lumped into three species based on reactivity, i.e., ARO1 (toluene, ethyl-benzene), ARO2 (m/p/o-xylene), and benzene. The atmospheric lifetimes of the three aromatic tracers against OH are 18 h, 4.2 h, and 3.9 days during the study period (13–25 October 2010), respectively. Due to the long atmospheric lifetime of benzene, it is more difficult to track and identify its sources; thus we do not explicitly discuss benzene in this study. We focus our analysis on reactive aromatics (toluene, ethyl-benzene, and m/p/o-xylene).

Initial and boundary conditions for chemical tracers are taken from GEOS-Chem (V9-02) 2∘ × 2.5∘ simulation (Bey et al., 2001). Anthropogenic emissions are from the MIX inventory for October 2010 (Li et al., 2017). MIX is a mosaic Asian anthropogenic emission inventory with the Multi-resolution Emission Inventory for China (MEIC) and several other emission inventories for other Asian countries. In addition to the MIX inventory, we also conduct sensitivity simulations using the Intercontinental Chemical Transport Experiment-Phase B (INTEX-B) emissions inventory (Zhang et al., 2009; Li et al., 2014), which was developed for the year 2006. We find that, compared to the in situ observations of aromatics, the simulation results using the INTEX-B emissions are better. The main reason for the simulation improvements is the emissions of aromatics in South Asia. Given the large uncertainties in the emissions of aromatics (e.g., Liu et al., 2012b), this result is not surprising. Since MEIC and INTEX-B inventories are developed by the same group, we replace MIX aromatic emissions outside China with INTEX-B data such that aromatic emissions in the model are consistent. The improvements of model simulations compared to in situ observations are shown in Fig. S1 in the Supplement. Since satellite observations are used to improve aromatic emissions (next section), using either MIX or INTEX-B emissions in this work gives the same conclusions. Biogenic VOC emissions are computed with the Model of Emissions of Gases and Aerosols from Nature (MEGAN) algorithm (v2.1; Guenther et al., 2012), and outdoor biomass burning emissions of CHOCHO and other species are based on Global Fire Emissions Database Version 4.1 with small fires (GFED4.1s; van der Werf et al., 2010; Andreae and Merlet, 2001; Lerot et al., 2010). Indoor burning CHOCHO emissions of India are computed using emission factors from Pettersson et al. (2011) and Li et al. (2014). Rural and urban population distributions of India for the year 2010 are used as spatial proxies (Balk et al., 2006; CIESIN, 2011, 2016). We adopt the energy consumptions for rural and urban inhabitants on the basis of the National Sample Survey Office of India (NSSO, 2012a, b). Compared with satellite-observed CO and NO2 VCDs, REAM performs reasonably well in the study region during October 2010 (Fig. S2 in the Supplement). For general model evaluations of REAM, we refer the readers to the papers cited early in this section.

We updated the INTEX-B emission inventory in South Asian countries through inverse modeling constrained by SCIAMACHY CHOCHO VCDs (next section). We run REAM simulations with the a priori and top-down emission inventories, and we compare the results with observations in Sect. 3.1 and 3.2, respectively. We find that some of the model low bias is likely due to emission underestimation. We further carried out three model sensitivity tests to calculate the contributions to surface aromatics from emissions over Tibet, other provinces of China, and South Asia (India and nearby regions). Each simulation is run with only the aromatic emissions from the corresponding region. The OH concentrations in each simulation are specified to the archived values of the full model simulation. The results for two sub-periods of period 2 are examined in Sect. 3.3.

Compared with SCIAMACHY data, REAM using the original emission inventories archived at the overpass time of SCIAMACHY underestimates CHOCHO VCDs in the populated regions of India (Fig. 2). This underestimation is especially significant in the Indo-Gangetic Plain located south of the Himalayas (Fig. 2c). We thus derive the top-down aromatic emissions for these regions constrained by SCIAMACHY CHOCHO data (Liu et al., 2012b; Alvarado, 2016).

First, we calculate the difference between observed (CCHOCHOSCIAMACHY, Fig. 2a) and modeled (CCHOCHOREAM, Fig. 2b) CHOCHO VCDs with original emissions (ΔCCHOCHO=CCHOCHOSCIAMACHY-CCHOCHOREAM, Fig. 2c). This discrepancy greatly exceeds the uncertainties of SCIAMACHY retrieval. We then discuss the potential reasons for the difference, i.e., primary emissions from biomass burning and secondary sources from isoprene, acetylene, ethylene, and aromatics (Fu et al., 2008; Liu et al., 2012b).

Biomass burning is often a major primary source of CHOCHO (Myriokefalitakis et al., 2008). GFED4.1s inventories, as well as fire hot spots observed by MODIS on board the Terra and Aqua satellites, indicate only a small number of fire occurrences during this period in South Asia, with the exception of crop residue burning in Punjab, an agricultural state in north India. The contribution to CHOCHO VCDs from outdoor biomass burning (Fig. S3a in the Supplement) differs greatly from that of ΔCCHOCHO (Fig. 2c), which is large over the industrialized Indo-Gangetic Plain. Simulated indoor burning contribution to CHOCHO VCDs is lower by a factor of about 15 than the CHOCHO VCD discrepancy between satellite retrieval and model simulation (Fig. S3b in the Supplement). The uncertainty of the indoor burning CHOCHO emissions mainly results from that of the emission factor. Even if we assume this uncertainty to be 300 %, indoor burning cannot explain the low bias of the simulated CHOCHO VCDs. Therefore, the large model underestimation of CHOCHO over the Indo-Gangetic Plain is unlikely due to outdoor biomass burning or indoor burning during our analysis period.

Direct anthropogenic emissions of CHOCHO are small (Volkamer et al., 2005; Stavrakou et al., 2009; Liu et al., 2012b). CHOCHO is produced primarily from the photochemical oxidation of biogenic compounds (e.g., isoprene and terpenes) and hydrocarbon released by anthropogenic activities (e.g., acetylene, ethylene, and aromatics) (Fu et al., 2008). Due to the long atmospheric lifetime of acetylene and ethylene, their contributions to CHOCHO concentrations are quite small in South Asia during October 2010. The most significant secondary sources of CHOCHO in South Asia are isoprene (Fig. 2d) and aromatics (Fig. 2e). Biogenic isoprene emissions depend on vegetation, sunlight, and temperature. The high isoprene contribution to CHOCHO VCDs is to the southeast of the Indo-Gangetic Plain, where CHOCHO VCDs are high in both the observations and model simulations. In comparison, aromatic oxidation dominates CHOCHO over the Indo-Gangetic Plain, where model underestimation is largest (Fig. 2).

We apply the approach by Liu et al. (2012b) to estimate the top-down emissions of aromatics based on SCIAMACHY CHOCHO VCDs. As found by Liu et al. (2012b), domain-wide inversion is impractical since model results correlate poorly with gridded satellite data, most likely reflecting the problems in the spatial distribution of a priori emissions. We therefore determine the emissions by inversion for each grid cell at the overpass time of SCIAMACHY as Liu et al. (2012b) and find similar results for India and nearby regions as Liu et al. (2012b) did for eastern China. The top-down biogenic isoprene emissions are essentially the same as the a priori emissions. However, the top-down anthropogenic emissions of aromatics (Fig. S4 in the Supplement) increase by a factor of 2–6. The improved model comparison with in situ observations will be discussed in the next section. One caveat with respect to the top-down emission estimate is that we have to assume that the speciation of aromatics in the a priori emission inventory is correct. Since the purpose of this work is to study transport pathways to the Tibetan Plateau on the basis of in situ observations, we examine lumped reactive aromatics (defined as the sum of toluene, ethyl-benzene, and m/p/o-xylene) in the model evaluation (next section). Satellite observations cannot be used for this purpose since CHOCHO VCDs over Tibet are below or around the detection limit.

The average of observed reactive aromatic surface concentration (59 ± 63 pptv) over the Tibetan Plateau is considerably lower than the values found for megacities of China, such as Beijing (8.04 ppbv) and Shanghai (5.2 ppbv) (Liu et al., 2012b). Higher aromatic levels were measured during period 1 (76 ± 39 pptv) and period 3 (169 ± 57 pptv) than in period 2 (26 ± 39 pptv). The model simulation using the a priori emissions in general compares poorly with the in situ observations (Fig. 3). The best performance is during the low-concentration period 2 when the model underestimates the observations by about a factor of 2. However, the relatively high correlation coefficient (R2=0.83) suggests that atmospheric transport and emission distribution are reasonably simulated. This is in sharp contrast to periods 1 and 3 when the model underestimates the observations by a factor of 7 with very low correlations between the model and the observations (slope = 0.14 and 0.15, R2=0.04 and 0.02 for period 1 and 3, respectively). We discuss the different reasons for the model performance for periods 1, 2, and 3 in the next three sections.

Figures 2 and S4 in the Supplement show that SCIAMACHY observations of CHOCHO suggest much higher industrial emissions of aromatics over the Indo-Gangetic Plain than the a priori emissions. We derive top-down emissions on the basis of SCIAMACHY CHOCHO VCDs (Sect. 2.4). Top-down emissions are higher than the a priori emissions by a factor of 2–6 over the Indo-Gangetic Plain, which is the upwind region of the Tibetan Plateau. Figure 4 shows the resulting improvement in the model simulation. The large underestimations of CHOCHO VCDs over the Indo-Gangetic Plain are corrected as expected (Fig. 4a). At the same time, in situ observations during period 2 are much better reproduced by the model with the slope increasing from 0.56 to 0.91 and a similar R2 value (0.66) (Fig. 4b). In contrast, model simulations for periods 1 and 3 are not improved using top-down emission estimates with low biases similar to the original model simulation. This indicates that the reasons for the discrepancies in periods 1 and 3 are probably not related to the uncertainties in emissions but could be linked to deficiencies in model transport in this area.

Observed and simulated reactive aromatic concentrations show large variabilities during period 2 (Fig. 4b). An investigation of these data shows that a major contributor is meteorology. Observed concentrations of reactive aromatics during 19–20 October are generally lower (6.6 ± 3.4 pptv) than those during 21–24 October (37 ± 45 pptv). The concentration difference during the two time periods is captured by model simulations with top-down emissions (Fig. 5). Analysis of WRF-simulated surface wind speed shows an increase by a factor of 2–4 from 19–20 October (Fig. 5a) to 21–24 October (Fig. 5b), corresponding well to increasing transport of aromatics from the Indo-Gangetic Plain.

To further analyze the difference between the two time periods, we conduct sensitivity simulations as described in Sect. 2.3. We compute the source attributions for emissions over Tibet, India and nearby regions, and China excluding Tibet (Fig. 6). During 19–20 October, reactive aromatics are due to Tibetan emissions. With the exception of one data point, the concentrations are ≤ 7 pptv. On 21 October, emissions from India and nearby regions become dominant while the concentrations are still low (7–21 pptv). During 22–24 October, however, emissions from India and nearby regions contribute much higher concentrations (10–137 pptv). The only exception is one data point sampled 30 km east of Lhasa, where about one-fifth of the population of Tibet resides. The contribution of emissions from India and nearby regions to this data point is ∼ 20 pptv, still much higher than during 19–20 October. The contribution of emissions from China (excluding Tibet) is negligible (∼ 1 %) for this period.

The rise of the Tibetan Plateau is a natural barrier for pollution transport (Fig. 1). Considering the high altitude of the Tibetan Plateau, we analyze the 300 hPa geopotential height field in order to understand the change of wind circulation over the region (Fig. 5). During 19–20 October, the upper troposphere shows a northward gradual pressure decrease, which does not promote near-surface forcing of trans-Himalaya transport (Fig. 5a). During 21–24 October, the presence of a southeastward-moving upper-tropospheric cut-off low system induces increasingly stronger surface wind from India to Tibet (Fig. 5b, Hoskins et al., 1985). The cut-off low system is a closed low-pressure system detached from the westerlies. It began to form on 21 October and started to dissipate on 24 October. Trans-Himalaya air mass flux in the lower atmosphere shows an increase by a factor of 2 to 5 (Fig. S5 in the Supplement). Accompanying this transport, large amounts of pollutants such as reactive aromatics analyzed here are transported to the Tibetan Plateau, leading to much higher surface concentrations.

The cut-off low system provides a more rapid and efficient pollutant transport pathway than transport pathways previously proposed by other studies, such as westerlies (Cong et al., 2015; Ji et al., 2015) and mountain-valley winds (Hindman and Upadhyay, 2002; Dumka et al., 2010). Compared to aromatics, BC is also subject to wet scavenging, which greatly reduces its transport efficiency by convection. In-cloud BC scavenging is due to cloud activation or ice nucleation and subsequent removal by precipitation, and below-cloud scavenging is due to collision with rain droplets (e.g., Taylor et al., 2014). During our analysis period, the cut-off low system and the associated precipitation are to the northwest of Tibet (Fig. S6 in the Supplement). Precipitation south of Tibet is weak, and thus the subsequent removal of BC during trans-Himalaya transport is limited.

To examine the sensitivity of trans-Himalaya transport to the distribution of emission sources, we redistribute the INTEX-B total aromatic emissions over China and other South Asia countries on the basis of the MIX BC emission distributions. We conduct a sensitivity simulation using the redistributed emissions and compare the results to the original simulation. The trans-Himalaya transport from South Asia clearly dominates, and it is strongly affected by the presence of a cut-off low system during our analysis period (Fig. S7 in the Supplement). Our analysis implies that BC transported in the presence of an upper-tropospheric cut-off low is potentially a major contributor to BC deposition to Tibetan glaciers.

Compared to period 2, model performance for periods 1 and 3 is very poor with severe low biases (Fig. 3). Transport deficiency appears to be the main problem. Figure 7 shows the histograms of observed and simulated surface wind speed for the three periods. The observed and simulated wind speed distributions are similar for period 2 (Fig. 7c). In comparison, the simulated wind speed distribution differs drastically for the other two periods (Fig. 7a and d).

The wind speed distributions are more similar between periods 1 and 2 in the observations than model simulations. The underestimation of wind speed in period 1 leads to slower transport of pollutants from the Indo-Gangetic Plain and consequently to a low bias in surface reactive aromatics in the model. Examination of the 300 hPa geopotential height field during 13–17 October of period 1 shows a weak trough northeast of Kazakhstan in CSFR reanalysis and WRF simulation results (Fig. S8 in the Supplement). A strong upper-tropospheric low-pressure system, akin to the cut-off low system of Fig. 5b, will induce stronger lower-troposphere wind circulation. The lack of radiosonde observations over the interior of the Tibetan Plateau to constrain the meteorological reanalysis is the plausible reason (Fig. S9 in the Supplement). The horizontal scale of the Rossby wave at northern midlatitudes is thousands of kilometers, which can be reasonably represented by the density of the existing radiosonde network. We hypothesize that the smaller-scale cut-off low system, not simulated in the reanalysis, is more likely the reason for the model–observation discrepancy during period 1. We resample surface wind speed of 23 October, when a cut-off low system leads to rapid trans-Himalaya transport in period 2 as analyzed in the previous section. At the same time of the day and location as the observations, the simulated wind speed histogram is in good agreement with the observations (Fig. 7b). The corresponding air mass flux across the Himalayas would have been much stronger in the presence of a cut-off low system (Fig. S10 in the Supplement).

During period 3, the observed wind speed histogram is skewed to very low wind speed (0–1 m s-1) compared to the simulations (Fig. 7d). Sampling bias to avoid locations with strong wind is a possible reason. Another reason is that these samples were taken at lower altitudes in valleys compared to higher altitudes in the other two periods. Inspection of Fig. 1 shows the complex terrain surrounding the valleys of period 3 sampling. Using high-resolution (∼ 1 km) terrain data from the US Geological Survey (USGS) Global 30 Arc-Second Elevation (GTOPO30) dataset, we find that the standard deviation of altitude in the 7 km × 7 km region centered at the corresponding observation location correlates well with the observed reactive aromatics with a R2 value of 0.55 (Fig. 8), which suggests that pollution transport is strongly enhanced by the effects of complex terrain. The horizontal resolution of 36 km used in this study is inadequate to simulate this effect. Model resolution as high as 1 km appears to be necessary to capture the observed feature, but the computational resource requirement will be exceptionally large for a global model such as that used for CFSR. Other issues related to complex terrains in this region were also discussed by previous studies (Maussion et al., 2011; Ménégoz et al., 2013; He et al., 2014; Kumar et al., 2015). The effects of complex terrain may have also affected the observations of periods 1 and 2 but to a smaller extent since the terrain variation is lower and sampling altitude is higher in those periods.