Pre-monsoon air quality over Lumbini , a world heritage site 1 along the Himalayan foothills

1 Key Laboratory of Tibetan Environment Changes and Land Surface Processes, Institute of 6 Tibetan Plateau Research, Chinese Academy of Sciences, Beijing 100101, China 7 2 University of Chinese Academy of Sciences, Beijing 100049, China 8 3 International Centre for Integrated Mountain Development (ICIMOD), Kathmandu, Nepal 9 4 Institute for Advanced Sustainability Studies (IASS), Potsdam 14467, Germany 10 5 Himalayan Sustainability Institute (HIMSI), Kathmandu, Nepal 11 6 State Key Laboratory of Cryospheric Science, Cold and Arid Regions Environmental and 12 Engineering Research Institute (CAREERI), Lanzhou 730000, China 13 7 Center for Excellence in Tibetan Plateau Earth Sciences, Chinese Academy of Sciences, Beijing 14 100085, China 15 8 Aryabhatta Research Institute of Observational Sciences (ARIES), Nainital, India 16 17


Introduction
The Indo-Gangetic Plain (IGP) stretches over 2000 km encompassing a vast area of land in northern South Asia: the eastern parts of Pakistan, most of northern and eastern India, southern part of Nepal and almost all of Bangladesh.The Himalayan mountains and their foothills stretch along the northern edge of IGP.The IGP region is among the most fertile and most intensely farmed region of the world.It is a heavily populated region with about 900 million residents or 12 % of the world's population.Four megacities -Lahore, Delhi, Kolkata and Dhaka -are located in the IGP region, with dozens more cities with populations exceeding 1 million.The region has witnessed impressive economic growth in recent decades but unfortunately it has also become one of the most polluted, and an air pollution "hotspot" of local, regional and global concern (Ramanathan et al., 2007).Main factors contributing to air pollution in the IGP and surrounding regions include emissions from vehicles, thermal power plants, industries, biomass and fossil fuel used in cooking and heating activities, agricultural activities, crop residue burning and forest fires.Air pollution gets transported long distances away from emission sources and across national borders.As a result, the IGP and adjacent regions get shrouded with a dramatic annual buildup of regional-scale plumes of air pollutants, known as atmospheric brown clouds (ABC), during the long and dry winter and pre-monsoon seasons each year (Ramanathan and Carmichael, 2008).Figure 1 shows monthly synoptic wind and mean aerosol optical depth during April-June 2013 over South Asia.Very high aerosol optical depth along the entire stretch of IGP reflects the severity of air pollution over large areas in the region.
Poor air quality continues to pose a significant threat to human health in the region.In a new study of global burden of disease released recently, Forouzanfar et al. (2015) estimated that in 2013 around 1.7 million people died prematurely in Pakistan, India, Nepal and Bangladesh as a result of air pollution exposure, nearly 30 % of global total premature deaths due to air pollution.Air pollution also affects precipitation (e.g., South Asian monsoon), agricultural productivity, ecosystems, tourism, climate and broadly socioeconomic and national development goals of the countries in the region (Burney and Ramanathan, 2014;Shindell et al., 2012;Ramanathan and Carmichael, 2008).It has also been linked to intensification of cold wave and winter fog in the IGP region over recent decades (Lawrence and Lelieveld, 2010, and references therein;Safai et al., 2009;Ganguly et al., 2006).Besides high levels of aerosol loading as shown in Fig. 1, IGP also has very high levels of ground-level ozone or tropospheric ozone (O 3 ) (e.g., Ramanathan and Carmichael, 2008), which is toxic to plant and human health and a major greenhouse gas (IPCC, 2013;Shindell et al., 2012;Mohnen et al., 1993).South Asia, in particular IGP, has been projected to be the most ozone-polluted region in the world by 2030 (Stevenson et al., 2006).The majority of crop loss in different parts of the world results from effects of ozone on crop health and productivity (Shindell et al., 2012).Burney and Ramanathan (2014) also reported a significant loss in wheat and rice yields in India from 1980 to 2010 due to direct effects of black carbon (BC) and O 3 .BC and O 3 are two key short-lived climate pollutants.Similarly, species like fine particles and carbon monoxide (CO) are potent to health damages by posing impacts upon the respiratory and cardiovascular system and even also to the climate system (Singh et al., 2017, and references therein).Because of the IGP's close proximity to the Himalaya-Tibetan plateau region, this once relatively clean region is now subjected to increasing air pollution transported from regions such as the IGP, which can exert additional risks to sensitive ecosystems in the mountain region (e.g., Lüthi et al., 2015;Marinoni et al., 2013;Duchi et al., 2011).However, air pollution transport pathways to Himalayas are still not yet fully understood.
Monuments and buildings made with stones are vulnerable to air pollution damage (Brimblecombe, 2003;Gauri and Holdren, 1981).The damage to the monuments and buildings could be in various forms like corrosion, soiling, abrasion and discoloration.For example, a recent study has reported that deposition of light absorbing aerosol particles (black carbon, brown carbon) and dust is responsible for the discoloration of Taj Mahal, a world-famous monument in India (Bergin et al., 2015).Lumbini, located near the northern edge of the central IGP, is famous as the birthplace of the Lord Buddha and thus a UNESCO world heritage site of outstanding universal value to humanity.Since the study area is renowned for its historical and archaeological significance, Lumbini is also getting worldwide attention for poor air quality in the region.There was no regular air quality monitoring in Lumbini at the time of our measurement campaign.
Through this study, we want to understand the level of air pollution, its diurnal characteristic and the influence of open burning on air quality in Lumbini.We carried out continuous measurements of ambient concentrations of key air pollutants (PM, BC, CO, O 3 ) and meteorological parameters during an intensive measurement period of 3 months (April-June) in the year 2013.These are the first reported pollutant measurements for Lumbini.A regional chemical transport model called Sulfur Transport and dEposition Model (STEM) was used to simulate the variations of meteorological parameters and air pollutants during the observation period to examine the extent to which a state-of-the-art, widely used air quality model is able to simulate the observations, as an indication for where there are still gaps in our knowledge and what further measurements and emissions dataset de- velopments are needed.Model-simulated regionally tagged CO tracers were used to identify emission source regions impacting pollutant concentration observed at Lumbini.Satellite data have also been used to understand the high-pollution events during the monitoring period.These measurements were carried out as a part of the SusKat-ABC international air pollution measurement campaign (Rupakheti et al., 2017) jointly led by the International Centre for Integrated Mountain Development (ICIMOD), Kathmandu, Nepal, and Institute for Advanced Sustainability Studies (IASS), Potsdam, Germany.
2 Experimental setup

Sampling site
The Lumbini measurement site (27 • 29.387 N, 83 • 16.745 E; elevation: ∼ 100 m a.s.l.) is located at the premise of the Lumbini International Research Institute (LIRI), a Buddhist library in Lumbini in the Rupandehi district.According to the national census conducted in 2011, the total population of Rupandehi district is about 900 000 with a population density of about 650 people km −2 , which is the fourth most densely populated district in the country.Over 130 000 tourists visited Lumbini in 2014 (http://www.tourism.gov.np/ne/).A local road (asphalt) lies about 200 m north of the sampling site and experiences intermittent passing of vehicles.About 25 km north of Lumbini the foothills begin, while the main peaks of the Himalayas are 140 km to the north.The remaining three sides are surrounded by flat plain land of Nepal and India.The site is only about 8 km from the Nepal-India border in the south.A three-storied 10 m tall water tower was used as the platform for the automatic weather station (AWS), while remaining instruments were placed inside a room near the base of the tower.An uninterrupted power back up was set up in order to assure the regular power sup-ply even during hours with scheduled power cuts during the monitoring period.Figure S1 in the Supplement shows the location of Lumbini, the Kenzo Tange Master Plan area of the Lumbini development project, the sampling tower and a brief discussion on the surroundings of the site.Outside of the Master Plan area lies a vast area of agricultural fields, village pockets, several brick kilns and cement industries.

Monitoring instruments
The summary of instruments deployed in Lumbini is presented in Table 1.All data were collected in Nepal standard time (NST), which is GMT + 05:45 h.PM 1 , PM 2.5 and PM 10 mass concentrations were monitored continuously with the EDM 164 (Grimm Aerosol Technik, Germany), which uses the light scattering at 655 nm to derive mass concentrations.Similarly, aerosol light absorptions at seven wavelengths (370, 470, 520, 590, 660, 880, 950 nm) were measured continuously with an Aethalometer (model AE-42, Magee Scientific, USA), averaging and reporting data every 5 min.It was operated at a flow rate of 5 L min −1 .No cutoff was applied for inlet; hence the reported concentration of BC is total suspended BC particles.As described by the manufacturer, ambient BC concentration is derived from light absorption at 880 nm using a specific mass absorption cross section.To obtain BC concentration in Lumbini, we used a specific mass absorption cross-section value of 8 m 2 g −1 for the 880 nm channel.A similar value has been previously used for BC measurement in the IGP (Praveen et al., 2012).To remove the filter loading effect, we used the correction method suggested by Schmid et al. (2006), which was also used by Praveen et al. (2012) for BC measurements at a rural site in the IGP.Surface O 3 concentration was measured continuously with an ozone analyzer (model 49i, Thermo Scientific, USA) which utilizes UV (254 nm wavelength) photometric technology to measure ozone concentration in ambient air.A CO analyzer (model 48i, Thermo Scientific, USA) was used to monitor ambient CO concentration.The ambient air was drawn through a 6 µm pore-size Savillex 47 mm filter at the inlet that removed the particles before sending the air into the CO and O 3 analyzers using a Teflon tube.The filters were replaced every 7-10 days depending on particle loading, based on manual inspection.The CO instrument was set to auto-zero at a regular interval of 6 h.Local meteorological parameters (temperature, relative humidity (RH), wind speed (WS), wind direction (WD), precipitation and global solar radiation) were monitored with an AWS (Campbell Scientific, Loughborough, UK), recording data every minute.

Regional chemical transport model
Aerosol and trace gas distributions were simulated using a regional chemical transport model.STEM, a 3-D Eulerian model that has been used extensively in the past to characterize air pollutants in South Asian region (Adhikary et al., 2010(Adhikary et al., , 2007)), was used to understand observations at Lumbini.The Weather Research and Forecasting (WRF) model (Skamarock et al., 2008)  The tracer version of STEM provides mass concentration of sulfate, BC (hydrophilic and hydrophobic), organic carbon (OC), sea salt (fine and coarse mode), dust (fine PM 2.5 and PM 10 ), CO (open burning and anthropogenic) and regionally tagged CO tracers.STEM domain size, resolution and projection are those of the WRF model.Details about the tracer version of STEM are outlined elsewhere (Kulkarni et al., 2015;Adhikary et al., 2007).Anthropogenic emission of various pollutants (CH 4 , CO, SO 2 , NO x , NMVOC, NH 3 , PM 10 , PM 2.5 , BC and OC) used in this analysis were taken from EDGAR HTAP v2 (http://edgar.jrc.ec.europa.eu/htap_v2/index.php?SECURE=123) for 2010.Annual emissions given in kg m −2 sec −1 at 0.1 • × 0.1 • resolution were converted to molecules cm −2 sec −1 and re-gridded to 25 km × 25 km resolution using four-point interpolation techniques available in the STEM emission preprocessor.The emissions were given a diurnal profile using previously used parameterization available in the preprocessor.Open biomass burning emissions on a daily basis during the simulated period were taken from data obtained from the FINN model (Wiedinmyer et al., 2011).As with the WRF model, STEM was run from 2 November 2012 to 30 June 2013 but data presented here are only during the intensive field campaign period.
3 Results and discussions

Meteorology
Hourly average time series of various meteorological parameters like precipitation in mm h −1 (Prec), temperature in • C (T ), relative humidity in %, WS in m s −1 and WD in degree during the monitoring period are shown in Fig. 2. Meteorological parameters were obtained with the sensors at the height of ∼ 12 m from the ground.Meteorology results from WRF model simulations have been used to indicate any significantly different air-mass type present during the measurement campaign after the meteorological observations malfunctioned.Precipitation data were derived from TRMM satellite (TRMM_3B42_007 at a horizontal resolution of 0.25 • ) from the Giovanni platform (http://giovanni.gsfc.nasa.gov/giovanni/)as the rain gauge malfunctioned during the sampling period.Precipitation data from TRMM (Fig. 2) show that Lumbini was relatively dry in the early portion of the measurement campaign, while the site did experience some rainfall events as the pre-monsoon edged closer to the monsoon onset.This lowered aerosol loading in the later half of the measurement campaign due to washout and less biomass open burning.Comparison of WRF model outputs with TRMM data shows that the model underpredicts rainfall through out the campaign.Average observed temperature for the sampling period until the sensor stopped working (on 8 May 2013, i.e., for 38 days of measurement) was 28.1 • C (minimum: 16.5 • C, maximum: 40 • C).Average temperature from the model, during same period, was 31 • C, with values ranging between 19 and 40 • C. As shown in Fig. 2, the model captures the variability of temperature and is mostly within the range of daily values.However, the model has a high bias and does not capture well daily minimum temperature values.The model data were interpolated to match the observation site's latitude, longitude and altitude for all variables discussed in this paper.In addition, the model does not show any large variation in temperature for the campaign period after the sensors stopped working.This insight will be useful to interpret pollution data later on.For the same period (until the sensor stopped working), the average (observed) RH was ∼ 50 % (ranging from 10.5 to 97.5 %) whereas the model showed the average RH to be ∼ 23 % with values ranging between 6 and 78 %.RH values are highly underestimated by the model but, as previously mentioned, the model does not show significant changes in RH during the measurement campaign after the observations stopped working.
Average observed wind speed during the study period was 2.4 m s −1 , with hourly values ranging between 0.03 and 7.4 m s −1 , whereas from the WRF model average wind speed was found to be 3.2 m s −1 (range: 0.06-11.1 m s −1 ).Diurnal variation of observed hourly average wind speed suggested that wind speeds were lower during nights and mornings while higher wind speed prevailed during daytime, with average winds > 3 m s −1 up to ∼ 3.3 m s −1 between 09:00 and 13:00 LT (Supplement, Fig. S2, lower panel).Highspeed strong winds (> 4 m s −1 ) were from the NW direction during the month of April and later switched to the almost opposite direction, i.e., SE direction, from the month of May onwards.The monthly wind rose plot using the data from both observation and modeling, where the difference in the pattern could be potentially due to the data resolution, is shown in Fig. S3.Comparing modeled wind direction prediction skills at the surface with one point measurement is not sufficient.However, in the absence of other measurements, we also show the comparison of wind direction as an indication of model performance over this region and not as model validation where a more high-resolution modeling and sensitivity analysis of model physics and chemistry may be required.Discrepancy on model results might have occurred due to various factors inherently uncertain in a weather prediction using a model.Additionally, air pollution transport occurs via elevated layers and is not limited to surface winds.We show NCEP/NCAR reanalysis plots at 850 hPa in Fig. S3 to illustrate the distinctly differing wind direction compared to the surface winds seen from observations as well as NCEP/NCAR reanalysis plot at 1000 hPa shown in Fig. 1.There are no upper wind measurement data nearby Lumbini to show model performance.Regardless, we believe that air quality model data are vital for understand- ing pollutant transport in an area where observation data are non-existent or incomplete.

General overview, PM ratios and influence of meteorology on pollution concentrations
Figure 3 shows hourly averaged time series of observed BC, PM 1 , PM 2.5 , PM 10 , O 3 and CO observed at Lumbini during the study period.Similar temporal behavior was shown by BC, particulate matter fractions (PM 1 , PM 2.5 and PM 10 ) and CO.The gap in the figure (for PM time series) is due to the power interruption to the instrument.BC concentrations during the measurement period ranged between 0.3 and 29.9 µg m −3 with a mean (±SD) value of 4.9 (±3.8) µg m −3 .BC concentrations in Lumbini during premonsoon months are lower compared to BC concentrations observed in the Kathmandu Valley because of high number of vehicles plying on the street, brick kilns and other industries in Kathmandu Valley (Sharma et al., 2012;Putero et al., 2015).The lowest concentration was observed during a rainy day (21-22 April) whereas the highest concentration was observed during a period of forest fire (detailed in Sect.3.3).For the entire measurement period, we found average (of hourly average values) PM 1 of 35.8 ± 25.6 µg m −3 (minimum-maximum range: 3.6-197.6µg m −3 ), PM 2.5 of 53.1 ± 35.1 µg m −3 (6.1-272.2µg m −3 ), PM 10 of 128.9 ± 91.9 µg m −3 (10.5-603.9µg m −3 ) and coarse-mode fraction of 75.7 ± 61.7 µg m −3 (1.9-331.8µg m −3 ).The coarse-mode (PM 10-2.5 ) fraction was ∼ 60 % of the PM 10 .The share of coarse-mode aerosol to PM 10 in Lumbini was higher than that observed in other sites in the IGP, such as Guwahiti, India (42 %) (Tiwari et al., 2017), and Dibrugarh, India (9-16 %) (Pathak et al., 2013), both in eastern IGP and Delhi (38 %) (Tiwari et al., 2015) in western IGP, indicating the higher contribution of coarse aerosols in Lumbini, likely lifted from soils from nearby agricultural fields and construction materials by stronger winds during pre-monsoon season.
Values of coarse-mode fraction, similar to Lumbini, have been reported by Misra et al. (2014) at Kanpur for dust dominated and mixed aerosols events.The share of BC in PM fractions was found to be ∼ 13 % in PM 1 , 9 % in PM 2.5 and ∼ 4 % in PM 10 but the correlation coefficients of BC with three PM fractions were found to be 0.89 (PM 1 ), 0.88 (PM 2.5 ) and 0.69 (PM 10 ), indicating the commonality in the sources of these pollutants.The contribution of BC in PM 1 was found to be ∼ 12 % in Kanpur during February-March (Kumar et al., 2016a), similar to Lumbini.Regarding the share of BC in PM 10 , the share observed in Lumbini (∼ 4 %) was similar to that observed over Varanasi (∼ 340 km south of our site) in central IGP (5 %) (Tiwari et al., 2016) and Dibrugarh in eastern IGP (∼ 5 %) (Pathak et al., 2013).Thus our results indicate that despite our station being located at the northern edge of the IGP along the foothills of the Himalayan range, the share of BC in PM is similar to that found in heavily polluted sites in the central and eastern IGP.
In Lumbini, the average (hourly) share of PM 1 in PM 2.5 , PM 1 in PM 10 and PM 2.5 in PM 10 was found to be ∼ 70, 34 and 47 %, respectively.Regarding other sites in IGP region, PM 2.5 / PM 10 ratios were reported to be 56 % in Kanpur (Snider et al., 2016), 60 % in Varanasi (Kumar et al., 2015),   (Tiwari et al., 2017), 90 % in Dribugarh (Pathak et al., 2013) and 62 % in Delhi (Tiwari et al., 2015), indicating local differences within IGP as well as suggesting that the influence of combustion sources at Lumbini is still lower compared to other locations in the Indian section of the IGP.A recent study (Putero et al., 2015) reported that the PM 1 / PM 10 during pre-monsoon of 2013 was found to be 0.39 in the Kathmandu Valley of Nepal.Lumbini has significantly lower vehicle emissions and population than the Kathmandu Valley yet the ratios are similar, indicating the importance of regional combustion sources in Lumbini for finer aerosols (PM 1 ) and soil-based emissions such as road dust in the Kathmandu Valley.Future studies will need to explore the emission sources around Lumbini in much greater detail.Lower PM 2.5 / PM 10 in Lumbini as compared to other regions mentioned earlier could be due to emissions from cement industries located within 15 km distance from the measurement site.
The observed 24 h average particulate matter concentrations (PM 2.5 and PM 10 ) were found frequently higher than the WHO prescribed guidelines for PM 2.5 (25 µg m −3 ) and PM 10 (50 µg m −3 ) with PM 2.5 exceeding 94 % and PM 10 of 85 % of the measurement period of 53 days in Lumbini.
Observed CO concentrations ranged between 124.9 and 1429.7 ppbv with an average value of 344.1 ± 160.3 ppbv.CO concentration observed in Lumbini is lower than that of Mohali, western India, where the average concentration was 566.7 ppbv during pre-monsoon season due to intense biomass and agro-residue burning over the region (Sinha et al., 2014).Temporal variation of CO concentrations is similar to that of BC, exhibiting very strong correlation (r = 0.9).Past studies have shown that the ratio of BC to CO depends upon multiple factors like site location, combustion characteristics (fuel and technology) at the sources and type of air mass (Girach et al., 2014;Pan et al., 2011;Zhou et al., 2009).Formation of the soot depends on the carbon-to-oxygen ratio of fuel whereas CO can also be produced naturally due to the oxidation of volatile organic compounds (Girach et al., 2014).Figure 4 shows the comparison of the average BC/ CO ratio (0.021) at Lumbini with that obtained from other sites.Please refer to Fig. S4 in the Supplement for the time series of BC/ CO ratio observed in Lumbini.We used the method described by Pan et al. (2011) to calculate the BC/ CO values.The ratio was calculated using the equation (BC − BC 0 )/(CO − CO 0 ), assuming the background values (BC 0 or CO 0 ) as 1.25 percentile of the data.The BC/ CO ratio in Lumbini is similar to that obtained at a suburban site, Pantnagar, in India (0.017) (Joshi et al., 2016) and in Maldives (0.017) (Dickerson et al., 2002), indicating the possibility of similar types of emission sources.However, the lower BC/ CO ratio obtained over megacities such as Beijing and Shanghai is due to the higher number of gasoline and diesel vehicles (Zhou et al., 2009).The ratios obtained at Lumbini are within the range of emission ratios from diesel used in transport sector (0.0013-0.055), coal (0.0019-0.0572) and biofuels (0.0087-0.0266) for domestic activities (Verma et al., 2010, and references therein), implying that BC and CO observed are from mixed sources.
The hourly averaged observed ozone concentration ranged between 1.0 and 118.1 ppbv with a mean value of 46.6 ± 20.3 ppbv during the sampling period.The 8 h maximum O 3 concentration exceeded WHO guidelines (of 100 µg m −3 ; (WHO, 2006) during ∼ 90 % of the measurement period.Our results clearly indicate that the current pollution levels in Lumbini are of great concern to the health of the people living in the region, including over a million visitors who visit Lumbini, and agro-ecosystems.
The relationship of wind speed to aerosol and gaseous pollutants in Lumbini is shown in Fig. S5 (Supplement).We were interested in studying the relationship between wind speed and the pollutants since the wind governs the horizontal dilution of the pollutants (Huang et al., 2012) and also the likelihood of lifting soil dust.Except ozone, all other pollutants exhibited negative correlation with wind speed.BC shows negative correlation (r = −0.42,P > 0.05) with wind speed, which is similar with other pollutants as well (as can be seen from the figure).Past studies have also reported a similar negative correlation of BC with wind speed over urban and sub-urban areas (Huang et al., 2012;Cao et al., 2009;Ramachandran and Rajesh, 2007;Sharma et al., 2002;Tiwari et al., 2013), indicating that the locally generated BC can accumulate in the atmosphere during lower wind speed conditions (Cao et al., 2009).Tiwari et al. (2013) also reported similar negative correlation (r = −0.45)during the pre-monsoon season over Delhi.In contrast, secondary pollutants like ozone exhibited a positive relation to the WS (r = 0.38, P > 0.05), indicating the location of precursor emission sources at some distance away from the measurement site.Solar radiation is one of the most important factors for production of ozone in the atmosphere (Naja et al., 2003).The correlation of hourly ozone concentration with solar radiation (not shown here) was found to be 0.41 (P > 0.05), whereas wind speed during the daytime only (06:00-18:00) showed very weak correlation of 0.02 (nonsignificant) with ozone, possibly indicating transport of precursors during nighttime.
Interestingly, the highest concentrations of all measured pollutants were obtained when the wind speed was less than 1 m s −1 .In a separate analysis (not shown here), we considered only the WS > 1 m s −1 and calculated the correlation coefficients to investigate the influence of regional emissions.We found the similar correlation values as previous when all WS values were considered (BC vs. WS = −0.41,CO vs. WS = −0.42,O 3 vs.WS = 0.29, PM 1 vs. WS = −0.40,PM 2.5 vs. WS = −0.38,PM 10 vs. WS = −0.33;all at P > 0.05).The correlation of WS (> 1 m s −1 ) with concentration of air pollutants indicates that air pollution over Lumbini is not only of local origin but also from other nearby regions.Past studies near this site have been focused on cities like Kathmandu (Sharma et al., 2012;Panday and Prinn, 2009;Putero et al., 2015) and Kanpur (Ram et al., 2010) and regions of IGP dominated by agro-residue burning (Rastogi et al., 2016;Sinha et al., 2014;Sarkar et al., 2013), all of which reported very high levels of pollution.Our study adds to the growing list of scientific observations in the IGP by providing data from the foothills of the central Himalayas.Very high aerosol loading is observed in South Asia during pre-monsoon, mostly over the IGP region (Supplement, Fig. S6).As this is the first study over an IGP site located in Nepal, pollution concentrations observed at Lumbini were compared with other sites in the region (Table 2).Different sites located at urban, semi-urban and remote locations were used for comparison to get a clear comparative picture of the situation at Lumbini amongst other locations in the region.Pre-monsoon seasonal average PM 2.5 concentration in Lumbini has been found to be lower than in the megacity Delhi (Bisht et al., 2015) and northwestern IGP (Sinha et al., 2014), possibly due to higher levels of emissions (from traffic and biomass burning, respectively) over those regions.In addition, average BC and CO concentrations in Lumbini fell between concentrations observed at rural sites (up to 6 times higher) and cities in the region (see Table 2), indicating that Lumbini, in a way, can still be considered a semi-urban location.The hourly average O 3 concentrations in Lumbini were found to be higher than in cities like Kathmandu (Putero et al., 2015) and Kanpur during the pre-monsoon season (Gaur et al., 2014).However from a mesoscale perspective, the hourly average O 3 concentrations were lower at Lumbini as compared to the base camp of Mount Everest region due to the uplift of polluted air masses (Marinoni et al., 2013), stratospheric intrusion (Cristofanelli et al., 2010) and even the regional or long-range transport of the air pollutants (Bonasoni et al., 2010) to the high-altitude site.
Regarding the monthly average concentration, the concentrations of all measured pollutants decreased as the premonsoon months advanced.The monthly average concentrations of the monitored species are shown in Fig. S7 along with the monthly fire hotspots over the region.Reduction in concentration (except PM) during the month of May (as compared to April) could be attributed to the fewer fire events during May as well as previously discussed washout by rainfall.Two peak pollution episodes observed during the first half of April and May are discussed in more detail in the next section.

Observation-model intercomparison
Chemical transport models provide insight to observed phenomena; however, interpretation has to take into account model performance before arriving at any conclusion.This section describes pollution concentrations simulated by the BC concentrations ranged between 0.4 and 3.7 µg m −3 with a mean value of 1.8 ± 0.7 µg m −3 for the period of 1 April-15 June 2013.The average model BC concentration was ∼ 2.7 times lower than the observed BC.Regarding PM 1 , PM 2.5 and PM 10 , the model-simulated average concentration was 12.3 ± 5.5 (0.9-41.7), 17.3 ± 6.7 (1.9-48.3)and 25.4 ± 12.9 (2.1-68.8)µg m −3 , respectively.The model estimated values were lower by a factor of 3 and 5, respectively, than the observed concentrations.The data show that the model needs much improvement in its ability to adequately predict observed aerosol characteristics at Lumbini given the input provided, e.g., emissions data.Since pollutant concentration is a function of emissions, transport, transformation and deposition, improvements in any of these areas would improve the model performance for this site.However, given observation insights by PM ratios, it seems that improvements are much needed in the emissions of primary aerosols.Current emissions (2010) do not account for trash burning, roadside dust and increasingly newer industries, especially emissions from cement factories that have popped up in recent years.We show sensitivity with emissions in a later section (Sect.3.3.2) in the vicinity of Lumbini, but emission improvements are needed beyond Lumbini, which is outside the scope of this paper.Average observed CO concentration was 255.7 ± 83.5 ppbv, ranging between 72.2 and 613.1 ppbv, with average model CO ∼ 1.35 times lower than observed.Time series comparison of modeled CO vs. observation is shown in Fig. 3. Apart from two peak episodes the model does a better job in predicting CO concentration over Lumbini.A previous study using STEM over Kathmandu Valley showed that the model was able to capture the annual BC mean value but completely missed the concentrations during pre-monsoon and post-monsoon period (Adhikary et al., 2007).Similar behavior is seen this time for CO where the model misses the peak values but reasonably captures CO concentration after mid-May when no biomass burning events are observed (model to observation ratio improves to 1.16).STEM CO performance can be significantly improved via better constraining emissions of open biomass burning as discussed in Sect.3.3.This activity is beyond the scope of this current paper although improvements are underway for all these sectors.

Diurnal variations of air pollutants and boundary layer height
In the emission source region, diurnal variations of primary pollutants provide information about the time-dependent emission activities (Kumar et al., 2016b).Figure 5 shows the diurnal variation of hourly averaged concentrations of measured pollutants during the sampling period.Primary pollutants like BC, PM and CO showed typical characteristics of an urban environment, i.e., diurnal variation with a morning and an evening peak.However, Lumbini data show higher concentrations in the evenings compared to morning hours.Elevated concentrations can be linked to morning and evening cooking hours for BC and CO.Emission inventories for the region show that the residential sector has significant contribution to BC and CO.However, an explanation for the elevated evening concentration compared to the morning is needed.Increase in the boundary layer height, reduction in the traffic density on the roads, absence of cooking activities during mid-day and increase in wind speed often contribute to the dispersion of pollutants, resulting in lower concentrations during the afternoon.Diurnal variation of wind direction (Supplement, Fig. S2, upper panel) shows the dominance of wind coming from the south (mainly during the month of May until mid-June).Morning and evening periods experienced the winds coming from the southeastern direction while the winds were predominantly from the southwestern direction during late afternoon.The increase in CO concentrations in the evening hours might be due to transport of CO from source regions upwind of Lumbini, which, along  (Kumar et al., 2016b).The PBL height (in meters) was obtained from the WRF model as observations were not available.The study period's average PBL height over Lumbini was ∼ 910 m (ranging between 24 and 3807 m, observed at 06:00 and 15:00, respectively).The daily average PBL height obtained from the model is compared with published values (Wan et al., 2017), as shown in Fig. 6, which indicate that the value is captured by our model during initial measurement period and overestimated in the months of mid-May onwards.As the pre-monsoon month advances, PBL height also increased.The monthly average PBL height was 799, 956 and 1014 m, respectively, during the month of April, May and (1-15) June.As presented in the figure, the monthly average diurnal variation also showed that the boundary layer height was at its maximum at 15:00 LT during each month, which coincides with the period of lowest concentration of the pollutants.

Influence of forest fires on Lumbini air quality
3.3.1 Identification of large-scale forest fire influence using in situ observations, satellite and model data Forest fires and agricultural biomass burning (mostly agroresidue burning on a large scale) are common over South Asia and the IGP region during pre-monsoon season.The northern Indo-Gangetic region is characterized by fires even during the monsoon and post-monsoon season (Kumar et al., 2016b;Putero et al., 2014).These activities not only influence air quality over nearby regions but also get transported towards high-elevation pristine environments like Mount Everest (Putero et al., 2014) and Tibet (Cong et al., 2015a,b).Thus, one of the main objectives of this study was to identify the influence of open burning on Lumbini air quality.Aver-age wind speed during the whole measurement period was 2.4 m s −1 .Based on these data, open fire counts within the grid size of 200 × 200 km centering over Lumbini were used for this analysis, assuming that the emissions would take a maximum period of 1 day to reach our monitoring site.Forest fire counts were obtained from the MODIS satellite data product Fire Information for Resource Management System (FIRMS). Figure 7 shows the daily average BC/ CO ratio, aerosol absorption Ångström exponent (AAE), which is derived from Aethalometer data (by calculating the negative slope of absorption at 370 and 950 nm vs. wavelength in log-log plot), and daily open fire counts within the specified grid.The green box in the figure is used to show two peak events (presented earlier in Fig. 3) with the elevated BC and CO concentrations observed during the monitoring period.The first peak was observed during 7-9 April and second peak during 3-4 May 2013.Two pollutants having biomass burning as the potential primary source, BC and CO, were taken in consideration.High AAE values during these two events are also an indication of the presence of BC of biomass burning origin (Praveen et al., 2012;Bergstrom et al., 2007;Kirchstetter et al., 2004), with the value being ∼ 1.6 for Lumbini.The chemical composition of TSP filter samples collected at Lumbini also showed higher concentration of Levoglucosan, a biomass burning tracer in Lumbini during the pre-monsoon season, compared to other seasons of the year (Wan et al., 2017).Wan et al. (2017) also reported that the higher correlation of K + with Ca 2+ and Mg 2+ indicates that dust is the main source of potassium in Lumbini.Contrary to our expectation, we could not observe any significant influence of forest fire within the specified grid of 200×200 km (the influence of local forest fire on the air quality over Lumbini was not observed).Therefore, a wider area, covering South and Southeast Asia, was selected for the forest fire count.Figure 8a and b show the active fire hotspots from MODIS over the region during the peak events, indicating the first peak could have occurred due to the forest fire over the eastern India region whereas the second peak was influenced by the forest fire over the western IGP.Moreover, in order to strengthen our hypothesis, we have utilized satellite data products for various gaseous pollutants like CO and NO 2 (Atmospheric Infrared Sounder (AIRS) for CO and  Ozone Monitoring Instrument (OMI) for NO 2 , both obtained from Giovanni platform).Figure 8c-h show the daytime total column CO before, during and after occurrence of two events (peaks), as stated earlier.The AIRS satellite with daily temporal resolution and 1 • × 1 • spatial resolution has been utilized to understand the CO concentration over the area.CO concentration over Lumbini during both of the peaks confirmed the role of open fires over the IGP region for elevated concentration of CO in Lumbini.To further strengthen our finding, HYSPLIT back-trajectory plots were used.Figure 8i and j represent the 6-hourly back trajectories for these two events only.However, the back trajectories (during both events) indicated that the air mass passed over the fire events in the northwestern IGP.We note that using back trajectories to identify source regions is also uncertain, as noted by Jaffe et al. (1997).Figure 8k shows model biomass CO peak coincident with observed CO.Although the magnitudes are significantly different, the timing of the peaks is captured well by the model.This, we believe, is due to the fact that satellitebased open fire detection has limitations because it cannot capture numerous small fires that are prevalent over South Asia but usually burn out before the next satellite overpass.More research is needed to assess the influence of these small fires on regional air quality.
In a separate analysis (not shown here), elevated O 3 concentrations during these two events were also observed.The average O 3 concentration before, during and after the events was found to be 46.2±20.3,53.5±31.1 and 50.3±20.9ppbv, respectively (Event I), and 54.8 ± 23.8, 56.7 ± 35 and 55.6 ± 13.4 ppbv, respectively (Event II).Average ozone concentration outside these events was found to be 46 ± 19 ppbv.Increased ozone concentrations during the high peak events have been analyzed using the satellite NO 2 concentration over the region and considering the role of NO 2 as a precursor for ozone formation.Daily total column NO 2 was obtained from the OMI satellite (data available at the Giovanni platform; http://giovanni.gsfc.nasa.gov/giovanni/)at the spatial resolution of 0.25 • × 0.25 • .Figure 9 shows the NO 2 column value before, during and after both events.Even for the NO 2 , maximum concentrations were observed during these two special events.It is likely that both local and regional pollution (transported from NW IGP region as indicated by synoptic wind in Fig. S8, Supplement) contributed to the elevated ozone levels.This remains a question to be investigated in future.

Identifying regional and local contribution
The WRF-STEM model has been used to identify the anthropogenic emission source region influencing the air quality over Lumbini.As previously explained, the model is able to capture the observed CO concentration when intense open burning events were not present.A recent study (Kulkarni et al., 2015) has explored the source region contribution of www.atmos-chem-phys.net/17/11041/2017/various pollutants over central Asia using similar technique.Figure 10a shows the average contribution from different regions to CO concentration over Lumbini during the whole measurement period.The major share of CO was from the Ganges Valley (46 %), followed by the Nepal region (25 %) and the rest of India (∼ 17.5 %).Contributions from other South Asian countries like Bangladesh and Pakistan were ∼ 11 % whereas China contributed ∼ 1 % of the CO concentration in Lumbini.Regarding the monthly average contribution, Ganges Valley's and Nepal's contributions were almost equal during the month of April (∼ 34 and ∼ 37 % respectively) but increased for the Ganges Valley during the month of May (∼ 44 %) and were reduced for Nepal (∼ 25 %) (Fig. S9, Supplement).
Figure 10b is the time series of percentage contribution to total CO concentration during the whole measurement period, showing different air masses arriving at 3-hourly intervals.During the whole measurement period, the majority of the CO reaching Lumbini was from the Ganges Valley (mainly the states of Punjab, Haryana, Uttar Pradesh, Bihar and West Bengal) region with the contribution sometimes reaching up to ∼ 80 %.Other Indian (central, south, east and north) regions also contributed significantly.Bangladesh's contribution of CO loading was seen only after mid-April, lasting for only about a week and after the first week of May.The contribution from Bangladesh was sporadic compared to other regions.The highest contribution from Bangladesh was observed after the first week of June with the arrival of monsoonal air mass.Pakistan also contributed to the CO loading significantly.Others region as mentioned in the figure covered the regions like Afghanistan, the Middle East, western Asia, East Asia, Africa and Bhutan.Contributions from these regions were less than 5 %.Contribution from China was not evident until the first week of June where a specific air-mass arrival shows contribution reaching up to 25 % of total CO loading.
A sensitivity analysis was performed for emission uncertainty in the model grid containing Lumbini.Lumbini and surrounding regions in the recent years have seen significant rise in urban activities and industrial activity and related emissions which may not be accurately reflected in the HTAP v2 emissions inventory.A month-long simulation was carried out with emissions from Lumbini and by switching off the four grids surrounding Lumbini, and another simulation with Lumbini and the surrounding four grids' emissions increased by 5 times the amount of HTAP v2 emissions inventory.The results are shown in Fig. 10c as percentage increase or decrease compared to model results using the current HTAP v2 emissions inventory.The black line shows the concentration as 100 % for the current HTAP v2 emissions inventory.Despite making Lumbini and the surrounding grids emissions zero, model calculation shows pollutant concentration on average is still about 78 % of the original value, indicating dominance of the background and regional sources compared to the local source in the model.Increasing emissions 5 times for Lumbini and surrounding four grids only increases the concentration on average by 151 %.Thus uncertainty in emissions is not a local uncertainty for Lumbini but rather for the whole region, which needs to be better understood for improving model performance against observations at Lumbini.

Does fossil fuel or biomass influence
Lumbini air?
The aerosol spectral absorption is used to gain insight into nature and potential source of black carbon.This method  enables to analyze the contributions of fossil fuel combustion and biomass burning contributions to the observed BC concentration (Kirchstetter et al., 2004).Besides BC, other light absorbing (in the UV region) aerosols are also produced in course of combustion, collectively termed organic aerosols (often also called brown carbon) (Andreae and Gelencsér, 2006).Figure 11 shows the comparison of normalized light absorption as a function of the wavelength for BC observed at Lumbini during cooking and non-cooking hours and also for both events.Our results are compared with the published data of Kirchstetter et al. (2004) and that observed over a village site of Project Surya in the IGP (Praveen et al., 2012) (figure not shown).We discuss light absorption data from two distinct times of the day.The main reason behind using data from 07:00-08:00 and 16:00-17:00 is these periods represent highest and lowest ambient concentration (Fig. 5).Also these periods represent cooking (07:00-08:00) and non-cooking (16:00-17:00) or high and low vehicular movement hours (Praveen et al., 2012).To understand the influence of biomass and fossil fuel we plotted normalized aerosol absorption at 700 nm wavelength for complete Aethalometer-measured wavelengths in Fig. 11.Kirchstetter et al. (2004) reported OC absorption efficiency at 700 nm to be zero.Thus we normalized measured absorption spectrum by 700 nm wavelength absorption.Since the Aethalometer does not provide 700 nm wavelength absorption values, we calculated the value using the absorption at nearby wavelengths and Ångström exponent following the methodology used by Praveen et al. (2012).Our results show that the normalized absorption for biomass burning aerosol is ∼ 3 times higher at 370 nm than that at 700 nm whereas fossil fuel absorption is about 2.6 times higher at the same wavelength.In addition, the curve obtained for the both events is inclined towards the published biomass burning curve.The normalized curve obtained during both cooking and noncooking period lies in between the standard curve of Kirchstetter et al. (2004).As shown in Fig. 11, the curve obtained for the prime cooking time is closer to the published curve on biomass burning whereas that obtained during the noncooking time is closer to the published fossil fuel curve.Similar results were also observed over the Project Surya village in the IGP region (Praveen et al., 2012;Rehman et al., 2011).This clearly indicates there is contribution from both sources, biomass and fossil fuel, to the observed BC concentration over Lumbini.
In order to identify fractional contribution of biomass burning and fossil fuel combustion to observed BC aerosol, we adopted the method described by Sandradewi et al. (2008).Wavelength dependence of aerosol absorption coefficient (b abs ) is proportional to λ −α , where λ is the wavelength and α is the absorption AAE.The α values ranges from 0.9 to 2.2 for fresh wood smoke aerosol (Day et al., 2006) and between 0.8 and 1.1 for traffic or diesel soot (references in Sandradewi et al., 2008).We have taken an α value of 1.86 for biomass burning and 1.1 for fossil fuel burning as suggested by previous literature (Sandradewi et al., 2008).Figure 12 shows diurnal variation of the biomass burning BC.Minimum contribution of biomass burning to total BC concentration was observed during 04:00-06:00 LT (only about 30 % of the total BC).As the cooking activities start in morning, the contribution of biomass BC starts to increase and reaches about 50 %.Similar pattern was repeated during evening cooking hours.Only during these two cooking periods, fossil fuel fraction BC was lower.Otherwise it remained significantly higher than biomass burning BC throughout the day.On average, ∼ 40 % of BC was from biomass burning whereas the remaining 60 % was contributed by fossil fuel combustion during our measurement period.Interestingly, this is the opposite of the contributions that were concluded by Lawrence and Lelieveld (2010).Lawrence and Lelieveld (2010) concluded that ∼ 60 % BC came from biomass vs. ∼ 40 % from fossil fuel, based on a review of numerous previous studies to be likely for studies on the outflows from South Asia during the winter monsoon.When we compared observed AAE with Praveen et al. (2012), we noticed that Lumbini values were lower than the Project Surya village site.This implies that Surya village center had a higher biomass fraction; additionally, it was observed absorption AAE exceeded 1.86 during cooking hours, which indicates 100 % biomass contribution.The difference is attributed to the fact that Lumbini sampling site is not a residential site like Surya, which can capture cooking influence efficiently.Further Lumbini sampling site is surrounded by commercial activities such as a local bus park, hotels, office buildings and industries and brick kilns slightly further away.Although the reason for this difference is not clear, it is an indication of the important role of diesel and coal emissions in the Lumbini and upwind regions.

Conclusions
Our measurements, a first for the Lumbini area, have shown very high air pollution at Lumbini.BC, CO, O 3 and particulate matter (PM 10 , PM 2.5 and PM 1 ) were measured during the pre-monsoon of 2013 at a regional site of the SusKat-ABC campaign.Average pollutant concentrations during the monitoring period were found to be 4.9 ± 3.8 µg m −3 for BC, 344.1 ± 160.3 ppbv for CO, 46.6 ± 20.3 ppbv for O 3 , 128.8±91.9µg m −3 for PM 10 , 53.14±35.1 µg m −3 for PM 2.5 and 36.6 ± 25.7 µg m −3 for PM 1 , which is comparable with other urban sites like Kanpur and Delhi in the IGP region.However, our study finds a higher fraction of coarse-mode PM in Lumbini compared to other sites in the IGP region.In addition, the BC/ CO ratio obtained in Lumbini was within the range of emissions from both residential and transportation sectors, indicating them as potential key sources of BC and CO and likely most of PM 1 in Lumbini.The diurnal variation of the pollutants is similar to that of any urban location, with peaks during morning and evening.However, our results show higher evening concentration compared to morning concentration values, further research is needed to explain this behavior.During our measurement period, air quality in Lumbini was influenced by regional forest fires as shown by chemical transport model and satellite data analysis.A regional chemical transport model, WRF-STEM, was used to understand observations.Intercomparison of WRF-STEM model outputs with observations showed that the model underestimated the observed pollutant concentrations by a factor of ∼ 1.5 to 5 but was able to capture the temporal variability.Model uncertainties are attributed mostly to uncertainties in meteorology and regional emissions as shown from sensitivity analysis with local emissions.Regionally tagged CO as air-mass tracers is employed in WRF-STEM model to understand the anthropogenic emission source region influencing Lumbini.Our analysis shows that the adjacent regions, mostly the Ganges Valley, other parts of India and Nepal, accounted for the highest contribution to pollutant concentration in the Lumbini.The normalized light absorption curve clearly indicated the contribution to BC in Lumbini from both sources: biomass as well as fossil fuel.On average, ∼ 40 % BC was found to be from the biomass burning and ∼ 60 % from fossil fuel burning.
Various improvements and extensions are possible in future studies.More reliable functioning of the AWS (temperature and RH sensor, rain gauge) would have allowed more in-depth analysis of the relationship between meteorological parameters and pollutants concentration.Continuous measurements of air pollutants throughout the year would allow for annual and seasonal variation study.Improvements in the model performance are much needed in its ability to simulate observed meteorology.Significant uncertainty lies with the regional emission inventory developed at national and continental scale vs. local bottom-up inventory and pollutant emissions from small-scale open burning not captured by satellites.There is a clear need to set up a continuous air quality monitoring station at Lumbini and the surrounding regions for long-term air quality monitoring.

Figure 1 .
Figure 1.Monthly synoptic wind (at 1000 hPa) for April, May and June 2013, based on NCEP/NCAR reanalysis data where the orientations of arrows refer to wind direction and the length of arrows represents the magnitude of wind (m s −1 ).Red square box in the figure (left) represents the location of Lumbini.Figures on the right side represent monthly aerosol optical depth acquired with the MODIS instrument aboard TERRA satellite.High aerosol loading can be seen over the entire Indo-Gangetic Plain (IGP).Light gray color used in the figure represents the absence of data.

Figure 2 .
Figure 2. Time series of hourly average observed (red) and model estimated (blue) meteorological parameters at Lumbini, Nepal, for the entire sampling period from 1 April to 15 June 2013.

Figure 3 .
Figure 3.Time series of the observed (red line) and model estimated (blue line) hourly average concentrations of BC, PM 1 , PM 2.5 , PM 10 , O 3 and CO at Lumbini, Nepal, for the entire sampling period from 1 April to 15 June 2013.

Figure 4 .
Figure 4. Comparison of black carbon concentrations to CO concentrations ( BC/ CO) ratios obtained for Lumbini with other sites.The red horizontal bar represents SD.

Figure 5 .
Figure 5.Diurnal variations of hourly average ambient concentrations of BC, PM 1 , PM 2.5 , PM 10 , O 3 and CO at Lumbini during the monitoring period (1 April-15 June 2013).In each box, the lower and upper boundary of the box represent the 25th and 75th percentile, respectively; the top and bottom of the whisker represent the 90th and 10th percentile, respectively; the mid-line represents the median; and the square mark represents the mean for each hour.

Figure 6 .
Figure 6.Daily time series of PBL height obtained from the model and reported values over Lumbini (obtained from Wan et al., 2017).The lower panel shows the monthly average diurnal variation of the PBL height.The square mark in each box represents the mean PBL height, the bottom and top of the box represent the 25th and 75th percentile and the top and bottom of the whisker represent the 90th and 10th percentile, respectively.

Figure 7 .
Figure 7. Time series of daily average BC/ CO ratio, absorption Ångström exponent (AAE) and fire counts acquired with the MODIS instrument on board the TERRA satellite for a 200 × 200 km grid centered at Lumbini.Two rectangular green boxes represent time of two episodes with high peaks in CO and BC concentrations as shown in Fig. 3.

Figure 8 .
Figure 8. Active fire hotspots in the region acquired with the MODIS instrument on Aqua satellite during (a) Event I (7-9 April) and (b) Event II (3-4 May).CO emissions, acquired with AIRS satellite, in the region 2 days before (3-5 April), during (7-9 April) and 2 days after (10-12 April) the Event I are shown in panels (c, e, g), respectively, while panels (d, f, h) show CO emissions 2 days before (1-2 May), during (3-4 May) and 2 days after (5-6 May) the Event II.Panels (i, j) represent the 6 h interval HYSPLIT back trajectories during Event I and II, respectively.Location of the Lumbini site is indicated by the red star in the panels (i, j).Observed CO vs. model open burning CO illustrating the contribution of forest fires during peak CO loading is shown in panel (k).

Figure 9 .
Figure 9. NO 2 total column obtained with OMI satellite over the region (a) before, (b) during and (c) after Event I.The panels (d-f) show NO 2 total column before, during and after Event II.

Figure 10 .
Figure 10.(a) WRF-STEM model estimated contributions of various source regions to average CO concentration in Lumbini for the sampling period; (b) time series of regionally tagged CO tracer during the whole measurement period using HTAP emission inventory; (c) percentage increase and decrease in CO concentration with different emissions scenario.

Figure 12 .
Figure 12.Diurnal variation of the fractional contribution of biomass burning to ambient BC concentration at Lumbini for the measurement period.In each box, the lower and upper boundary of the box represent the 25th and 75th percentile, respectively, and the top and bottom of the whisker represent the 90th and 10th percentile, respectively.The mid-line in each box represents the median while the square mark represents the mean for each hour.

Table 1 .
Summary of instruments deployed during monitoring in Lumbini.

Table 2 .
Comparison of PM 2.5 , BC, CO and O 3 concentrations at Lumbini with those at other sites in South Asia.

Table 3 .
Intercomparison of observed and model-simulated hourly average concentrations of air pollutants during the measurement campaign period.Units: BC and PM in µg m −3 and CO in ppbv.