Figure 3 shows hourly averaged time series of observed BC, PM1, PM2.5, PM10, O3 and CO observed at Lumbini during the study period. Similar temporal behavior was shown by BC, particulate matter fractions (PM1, PM2.5 and PM10) 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 µgm-3 with a mean (±SD) value of 4.9 (±3.8) µgm-3. BC concentrations in Lumbini during pre-monsoon 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) PM1 of 35.8±25.6 µgm-3 (minimum–maximum range: 3.6–197.6 µgm-3), PM2.5 of 53.1±35.1 µgm-3 (6.1–272.2 µgm-3), PM10 of 128.9±91.9 µgm-3 (10.5–603.9 µgm-3) and coarse-mode fraction of 75.7±61.7 µgm-3 (1.9–331.8 µgm-3). The coarse-mode (PM10–2.5) fraction was ∼60 % of the PM10. The share of coarse-mode aerosol to PM10 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 PM1, 9 % in PM2.5 and ∼4 % in PM10 but the correlation coefficients of BC with three PM fractions were found to be 0.89 (PM1), 0.88 (PM2.5) and 0.69 (PM10), indicating the commonality in the sources of these pollutants. The contribution of BC in PM1 was found to be ∼12 % in Kanpur during February–March (Kumar et al., 2016a), similar to Lumbini. Regarding the share of BC in PM10, 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 PM1 in PM2.5, PM1 in PM10 and PM2.5 in PM10 was found to be ∼70, 34 and 47 %, respectively. Regarding other sites in IGP region, PM2.5/PM10 ratios were reported to be 56 % in Kanpur (Snider et al., 2016), 60 % in Varanasi (Kumar et al., 2015), 57 % in Guwahiti (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 PM1/PM10 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 (PM1) 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 PM2.5/PM10 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 (PM2.5 and PM10) were found frequently higher than the WHO prescribed guidelines for PM2.5 (25 µgm-3) and PM10 (50 µgm-3) with PM2.5 exceeding 94 % and PM10 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 ΔBC/ΔCO ratio (0.021) at Lumbini with that obtained from other sites. Please refer to Fig. S4 in the Supplement for the time series of ΔBC/ΔCO ratio observed in Lumbini. We used the method described by Pan et al. (2011) to calculate the ΔBC/ΔCO values. The ratio was calculated using the equation (BC-BC0)/(CO-CO0), assuming the background values (BC0 or CO0) as 1.25 percentile of the data. The ΔBC/ΔCO 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 ΔBC/ΔCO 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 O3 concentration exceeded WHO guidelines (of 100 µgm-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 (non-significant) 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 ms-1. In a separate analysis (not shown here), we considered only the WS>1 ms-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, O3 vs. WS=0.29, PM1 vs. WS=-0.40, PM2.5 vs. WS=-0.38, PM10 vs. WS=-0.33; all at P>0.05). The correlation of WS (>1 ms-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 PM2.5 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 O3 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 O3 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 pre-monsoon 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.

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 WRF-STEM model. A comparison of model calculated pollutant concentration along with the minimum and maximum concentrations of various pollutants (with observation) is shown in Table 3. The model-based concentrations used here are values outputted for every third hour of the day (actual computation is carried out every 15 min). BC concentrations ranged between 0.4 and 3.7 µgm-3 with a mean value of 1.8±0.7 µgm-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 PM1, PM2.5 and PM10, 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) µgm-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.

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 with the local emissions, get trapped under reduced planetary boundary layer (PBL) heights. Ozone concentration was lowest in the morning before the sunrise and highest in late afternoon around 15:00 after which concentrations started declining, exhibiting a typical characteristic of a polluted urban site. Photo-dissociation of accumulated NOx reservoirs (like HONO) provides sufficient NO concentration, leading to the titration of O3 and resulting in minimum O3 just before sunrise (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.

Forest fires and agricultural biomass burning (mostly agro-residue 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. Average wind speed during the whole measurement period was 2.4 ms-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 ΔBC/ΔCO 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 Ca2+ and Mg2+ 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 NO2 (Atmospheric Infrared Sounder (AIRS) for CO and Ozone Monitoring Instrument (OMI) for NO2, 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 satellite-based 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 O3 concentrations during these two events were also observed. The average O3 concentration before, during and after the events was found to be 46.2±20.3, 53.5±31.1 and 50.3±20.9 ppbv, 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 NO2 concentration over the region and considering the role of NO2 as a precursor for ozone formation. Daily total column NO2 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 NO2 column value before, during and after both events. Even for the NO2, 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.

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 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.