Variability of Anthropogenic Gases: Nitrogen Oxides, Sulfur Dioxide, Ozone and Ammonia in Kathmandu Valley, Nepal

ABSTRACTKathmandu Valley is one of the largest and most polluted metropolitan regions in the Himalayan foothills. Rapidly expanding urban sprawl and a growing fleet of vehicles, and industrial facilities such as brick factories across the valley have led to conditions where ambient concentrations of key gaseous air pollutants are expected to exceed Nepal’s National Ambient Air Quality Standards (NAAQS) and World Health Organization (WHO) guidelines. In order to understand the spatial variation of the trace gases in the Kathmandu Valley, passive samples of SO2, NOx, NO2, NH3, and O3 were collected simultaneously from fifteen locations between March and May 2013. A follow-up study during two separate campaigns in 2014 sampled these gases, except ammonia, one site at a time from thirteen urban, suburban and rural stationary sites. In 2013, urban sites were observed to have higher weekly averaged NO2 and SO2 (22.4 ± 8.1 μg m–3 and 14.5 ± 11.1 μg m–3, respectively) than sub-urban sites (9.2 ± 3.9 μg m–3 and 7.6 ± 2.8 μg m–3, respectively). Regions located within 3 km of brick factories had higher SO2 concentrations (22.3 ± 14.7 μg m–3) than distant sites (5.8 ± 1.1 μg m–3). Higher O3 (108.5 ± 31.4 μg m–3) was observed in rural locations compared to urban sites (87.1 ± 9.2 μg m–3), emphasizing the importance of meteorological factors and precursor species for ozone production and titration. Parallel to previous studies, these results suggest that ground-level O3, as its levels frequently exceeded guidelines throughout the sampling periods, is an important concern throughout the valley. NH3 near polluted rivers and SO2 around brick factories are also important pollutants that need more intensive monitoring, primarily due to their importance in particulate matter formation chemistry.

According to World Health Organization (WHO), approximately 3.7 million deaths out of total global mortality were attributed to outdoor air pollution in 2012 alone (WHO, 2014), and the mortality rates nearly double when taking into account impacts from indoor air pollution. In 2012, about 89% of the premature deaths due to the ambient air pollution occurred in low and middle income countries from illnesses such as stroke, heart disease, lung cancer and both chronic and acute respiratory diseases (WHO, 2014).
The United States Environmental Protection Agency (US EPA) reports that the major constituents of atmospheric air pollutants include particulate matter and anthropogenic gases (EPA, 2012). Among anthropogenic gases, the US EPA has designated tropospheric ozone, carbon monoxide, sulfur dioxide and oxides of nitrogen as criteria pollutants due to their significant potential to affect human health and the environment. Due to such pollutants, deteriorating air quality has been observed in many cities in Asia, Europe and North America (Fang et al., 2009, Pascal et al., 2013, Parrish et al., 2011, and notably in rapidly developing regions or countries. High levels of air pollutants have been observed even in remote regions like the Himalayas and Tibetan Plateau (e.g., Marinoni et al., 2010;Lüthi et al., 2015) which are still relatively poorly sampled regions of the world.
The primary reasons for poor air quality in fast growing cities are emissions from rapidly increasing yet poorly maintained vehicles (Rakowska et al., 2014), and industries with no or limited pollution control (Lim et al., 2010) to meet the needs of growing population and demands, as well as open burning of municipal waste and agro-residues which are common in developing countries (Shakya et al., 2010). This poses a significant threat to human health (Gurung and Bell, 2013), environmental degradation (Zhang et al., 2012), and economic stability (Dhimal, 2009).

Air pollution in Kathmandu Valley
Kathmandu Valley is surrounded by mountains creating a distinctive bowl shaped area which is conducive for physically limiting ventilation and vertical movement of air in the valley, as well as leading to wintertime thermal inversions trapping pollutants near the surface Sapkota and Dhaubhadel, 2002). Consequently, air pollutants emitted from natural and anthropogenic sources build up and remain within the valley with little to no exchange of low level cold air with warm air sitting above the valley (Panday et al., 2009: Sapkota andDhaubadel, 2002), making the valley one of the most polluted regions in Nepal.
Previous studies have identified Kathmandu's urban sprawl, rapidly increasing vehicle fleet, and expanding industries as contributors to poor air quality within the valley and boarder surrounding region in the Himalayan foothills (Shah and Nagpal, 1997). Particulate pollution has been the main concern in Kathmandu Valley with PM 2.5 levels regularly exceeding the WHO guidelines in Kathmandu (Aryal et al., 2009). Trace gases, such as ozone, have also been observed to be at very high levels in Kathmandu Valley (Pudasainee et al., 2006). Previous studies (Byanju et al., 2012;Pradhan et al., 2012) have highlighted the trace gases related with anthropogenic emissions such as NOx, SO 2 , and O 3 in Kathmandu Valley.
In the Kathmandu Valley, there are many unregulated sources of air pollutants such as extensive fossil fuel combustion in nearly 250,000 backup power generators of various sizes, coal used in more than 100 brick factories, open burning of municipal waste, landfill sites, unpaved roadways, poorly maintained motor vehicles, and other high-temperature combustion processes in industrial operations (Dhimal, 2009;Pradhan et al., 2012).
Only a small number of emissions control strategies or regulations exist to ameliorate the air quality concerns in the Kathmandu Valley, but due to weak enforcement of existing regulations, many of these sources continue to emit pollutants unabated.

Health effects of gaseous pollutants
Degraded air quality is linked to a variety of health effects, and that these exposures trigger both direct health end points (inflammation, toxic chemical accumulation) as well as act as triggers for longer term non communicable diseases (cardiovascular and pulmonary diseases).
Several acute and chronic effects on human health have been associated with breathing NOx (Chen et al., 2012), SO 2 (Kan et al., 2010) and O 3 (Jhun et al., 2014). A wide range of health effects are observed ranging from upper respiratory irritation to chronic respiratory disease, aggravated asthma attacks, acute respiratory infections in children and chronic bronchitis in adults, heart disease and lung cancer, premature mortality and reduced life expectancy (WHO, 2005). Exposure to increased levels of NO 2 and O 3 can also induce oxidative stress that triggers a variety of health problems such as lung inflammation, shortness of breath and aggravated asthma, emphysema, and chronic bronchitis (EPA, 2012).
NO 2 has adverse health effects in both long and short term exposure timeframes, and asthmatics have been shown to be more sensitive to NO 2 exposure. NO 2 is strongly related to PM, as both are often emitted from the same combustion sources which makes it is difficult to differentiate the effects of NO 2 from those of other pollutants in epidemiology studies (WHO 2006). Studies have shown that a strong correlation exists between NO 2 and fine and ultrafine particulate matter, inorganic acids, elemental and organic carbon and therefore this gas can be considered as a reasonable tracer of the complex gas-particle mixture that originates from traffic (Seaton and Dennekamp, 2003). A multi-city study in Europe has found that the effect of PM on daily mortality was greater in areas with high NO 2 levels (Samoli et al 2003). Short term exposure studies have indicated that daily average concentrations of NO 2 are significantly associated with increased overall, cardiovascular, and respiratory mortality. For example, after adjusting for the effects of PM, a meta-analysis report shows a 0.9% increase per 24 ppb (45.9 µg/m 3 ) of NO 2 (24hr mean) as an effect estimate for all-cause mortality (Stieb et. al., 2002). In addition, a 10 µg/m 3 increase in 24hr mean NO 2 level has been found to increase hospital admissions due to respiratory causes by 0.4-0.5% (Bylin and Forsberg, 2009). Other studies have suggested that long-term exposure of oxides of nitrogen (i.e. NO 2 and NO) were associated with an increased risk of all-cause mortality (Felleul et., al, 2005, Hoek et. al., 2002, although most of the association was attributed to PM pollution. Other long-term effects of increased concentrations of NO 2 include an observed association with childhood cancer and lung cancer in adults, association with decreased lung function in children and in adults, and association with an increased incidence of asthma in children (WHO 2006) Acute and chronic exposure to O 3 has been associated with several health outcomes such as inflammation and reduction of pulmonary function. For short-term exposure, WHO reports that there is a significant increase in mortality observed above 50-70µg/m 3 of O 3 measured at 8 hour average (WHO, 2008). WHO also reports that, unlike NO 2 , the health effects of O 3 occur independently without being masked by PM. at least in short term exposures. For example , Bell et al (2005) and Levy et al., (2005) showed a significant association between O 3 and short-term mortality that was not substantially altered by exposure to other pollutants (such as PM) and factors such as meteorological parameters. Increases in total mortality have been observed at a concentration as low as 75 μg/m 3 (1-hour mean) (Gryparis et al., 2004). Other studies have used hospital admission attributed to respiratory tract infections and exacerbation of chronic airway diseases as health end-points to find health effect associations to acute O 3 exposure. For example, an increase of 100 μg/m 3 O 3 level was associated with admissions for pneumonia with a 2-day lag (RR = 1.14, 95% CI 0.94-1.38) and for COPD with a 1-day lag (RR = 1.17, 95% CI 0.86-1.60) amongst and elderly population in Alabama (Schwartz, 1994a), Michigan (Schwartz 1994b) and Minneapolis (Schwartz 1994c (Ostro et al. 1999). Consistent results were observed in long term exposure studies, where Jerrett et al (2005) found a 2% estimated increase in relative risk of death from respiratory causes that was associated with 10µg/m 3 increment in O 3 concentration.
As a primary irritant, SO 2 can have severe health effects, both short-term and long-term.
In general, chronic exposure to SO 2 is associated an overall increase in morbidity and mortality (Zhang et al 2000). Several health effects of chronic exposure include: wheezing, mild dyspnoea, persistent cough, and chronic obstructive pulmonary disease (COPD) (Wichmann andHeinrich 1995, andJammes 1998). Exposure to SO 2 has also been associated with cardiovascular abnormalities including decrease in heart rate variability (Tunnicliffe et al. 2001). In two European cities (Paris and Milan), there was a significant 4% increase associated with hospital respiratory admissions in associated with annual mean SO 2 of 50 µg/m 3 (Dab et al 1996). And the Harvard Six cities study showed that increased mortality was observed in cities with higher SO 2 levels (Dockery et al 1993). Increases in mortality have been associated with increased proportions of ambient SO 2 levels. The elevated death rate is due to respiratory and pulmonary diseases, along with cardiovascular and cerebrovascular diseases, including coronary and ischemic heart diseases and atherosclerotic diseases (Chang et al 2003). Short term (less than 24 hour) SO 2 exposure studies report that health effects such as reductions in Forced Expiratory Volume in 1 sec (FEV1), increases in specific airway resistance, and symptoms such as wheezing or shortness of breath (Linn et al. 1984). Long-term exposure to lower concentrations of SO 2 causes throat irritation and bronchoconstriction, particularly among individuals with asthma (Kampa & Castanas, 2008). Although the main target system for exposure and toxic action is the respiratory system, SO 2 can also cause severe irritation to other body parts such as the eye and skin from direct contact (Witek et al 1985a).
In addition to their direct impact on health, such gaseous pollutants also cause harm to health by contributing to the formation of secondary aerosols such as sulfate, nitrate, and ammonium that were found to be the main inorganic constituent of PM at an urban location in Kathmandu (Shakya et al., 2010). Health effects of ammonia are often indirect through contribution to ammonium ion formation which increased the health burden attributed to particle matter formation. Adverse health effects are often strongly associated with exposure to particulate matter (PM) (WHO, 2005), and it is one of the main health risk in the Kathmandu Valley .

Sources of gaseous pollutants
The main anthropogenic sources of these gases are mobile and stationary combustion sources, such as fossil fuel combustion. Oxides of nitrogen, such as NO 2 , are formed when NO (primary pollutant emitted from combustion activity) rapidly reacts with oxidizing agents such as O 3 or radicals in the surrounding atmosphere forming NO 2 . Tropospheric ozone (i.e. groundlevel ozone) is formed by a series of reactions involving NO 2 and Volatile Organic Compounds (VOCs), with the catalytic involvement of ultraviolet radiation from sun light (Venkataramani et al., 2010). Anthropogenic SO 2 emission results from the combustion of sulfur containing fossil fuels such as coal, heavy oils and the smelting of sulfur containing ores (Kampa & Castanas, 2008). Although NH 3 is most present from natural sources as it is emitted by microorganisms involved in the decay of animal matter and is often a byproduct of agricultural activities, it can be used as an indicator for poor sewage treatment systems that can pollute rivers and streams.
Industries such as coke and ammonia production factories also contribute to anthropogenic emissions of NH 3 in the atmosphere

Air quality standards in Nepal
Nepal has implemented a system of National

Passive sampling
Air quality monitoring and sampling is often conducted by either passive or active sampling techniques. Unlike active sampling which requires the use of a pump to actively pull air through the collection media; passive sampling does not require active air movement from a pump. Rather, gases are captured by a diffusion process through a static air layer or permeation through membrane in accordance with Fick's first law of diffusion (Salter, 2005). The basic principle on which passive samplers operates is of molecular diffusion, whereby the molecules of gas diffusing from the region of the high concentration (i.e. the surrounding atmosphere) to the region of the low concentration (i.e. absorbent pad of the sampler). With efficient reliability, the filters used in passive samplers are built for single-use application designed to absorb or attach to a specific pollutant over sampling time periods ranging from weeks to months depending on study timeframe.
Passive samplers can be used to monitor gaseous pollutants such as O 3 , NH 3 , SO 2 and NO 2 . In recent years, various types of passives samplers such as Ivl-Sweden, Gradko-UK, and Ogawa-Japan have been introduced into the global market. Ogawa passive samplers have been widely used and validated for ambient trace gas monitoring (Meng et al., 2010;Roadman et al., 2003). As a badge type passive sampler manufactured by the Ogawa Incorporated, Japan, the Ogawa sampler is comprised of two chambers (Ogawa & Co., 2014). In each chamber of the sampler, the setup of the sampler starts at the innermost position with the pad and pad retaining ring, stainless steel screen along with coated collection filter pads which progress outwards to the diffuser end cap (Figure 1.) The Ogawa filter pad's impregnated with pollutant specific absorbing reagents was used for sampling the gases. The filter pads are coated with a different chemical absorbent depending upon which pollutants to be monitored were loaded on the sampler (Table 1).  A particular challenge in conducting measurements in Kathmandu is the lack of operational ambient air monitoring stations, which have been largely nonfunctional since 2008 mainly due to frequent power outages and the lack of financial power to provide the necessary maintenance needed for the monitoring equipments (Dhimal, 2009). Routine online monitoring techniques for criteria gaseous pollutants are expensive and require regular maintenance which is often not available within the country. Distributed grid electricity is often interruptedas much as 12-14 hours per day without electricity, particularly in the winterwhich greatly restricts the measurement capacity of investigators. Because of these logistical and financial limitations, adopting a passive sampling measurement method is a useful option for collecting samples in developing countries such as Nepal. Passive sampling technique technique has proven reliable and effectively used in developing African (Adon et al., 2010) and south Asian countries (Byanju et al., 2012;Behera et al., 2015) to collect trace gaseous pollutants from ambient air because they are low cost, require little training and oversight, and need almost no research infrastructure support in the field. Therefore, these samplers can be effective for a developing country like Nepal which still lacks effective continuous monitoring (Pradhan et al., 2011). Furthermore, knowledge about the temporal and spatial distributions of these trace gases gained through such passive sampling techniques can help assess the emission sources of these gases, and hence improve the understanding of air quality and its health effects in Kathmandu Valley. By utilizing passive sampling techniques, this paper has attempted to provide a broad and comprehensive air quality analysis from many sites including rural, suburban, urban, and roadside sites across the c) Are there any specific known sources to these gases near sampling regions?

Hypothesis Statement
Weeklong passive sampling in Kathmandu Valley will capture ambient levels of pollutants that exceed Nepal's ambient standards and recommended levels that may have negative implications on the health outcomes of the population residing within the valley. In addition, the concentration and dispersion of these gaseous pollutants in the valley depends on factors such as the presence of local and distant emission sources, and topographic and meteorological conditions that will cause apparent spatial and temporal variations.

Objectives
My objective through this work is: (1) to make a quantitative analysis of the distribution of gaseous pollutants across the valley characterized by different emission sources, and (2) to compare our findings with previous studies conducted in the valley and (3) to assess influence of meteorology in the region on the temporal variability and spatial distribution of these gases, including differences at sites upwind and/or downwind of three major cities -Kathmandu   (1) Site type is based on local GIS data from rds.icimod.org. Furthermore, this classification is supported by approximate distance (in km) from Kathmandu city's core ring road. Urban < 3km; 3km < Suburban < 6km; and Rural > 6km. Land use type adopted from GIS data viewer (rds.icimod.org).
(2) Region classification is based on geographic location with respect to the main ring road surrounding Kathmandu city. Sites within the core ring road are classified as central regions. Kathmandu Valley includes three districts: Kathmandu, Bhaktapur, and Laltipur. (3) Coordinates and altitude, in meters, adopted from Google Earth and IASS POTSDAM database. During the following year in 2014, as part of a field study looking at traffic officers' personal exposure to air pollutants, 72 week-long measurements of the same gaseous pollutants, with the exception of NH 3 , were collected between February-April and July-September 2014 from 13 sites (Appendices: Measurement from 2014). The study design was particularly focused on studying personal exposure among traffic officers, rather than addressing regional background measurements in the valley as the 2013 dataset. Furthermore, unlike 2013 measurements, most of the samples from 2014 were taken within close proximity to major roadways, at least in the case of the urban on-road sampling sites. Therefore, due to lack of extensive regional measurement dataset and insufficient data to make direct comparison to the 2013 dataset, I have moved the 2014 data into an appendix (Appendix: I-V), and hence not discussed here.  (Giri et al., 2006). The climate is influenced by the South Asian summer monsoon as it receives up to 90 % of its annual rainfall during the three (June, July and August) summer months (Panday, 2006). The relative humidity reaches more than 80% in monsoon season than in the dry season (50%) in Kathmandu Valley (Giri et al., 2008).

SusKat-ABC Campaign
The valley has a population of 3 million and maintains a 4 percent per year population growth rate (Muzzini and Aparicio, 2013), and is the largest metropolitan area in the Himalayan foothills.

Passive sample collection
All samplers in this study were deployed by trained individuals and affixed under rain shelters. The samplers were exposed to ambient air at each site for 5-7 days before being collected and replaced. In addition, field blanks were also collected in the same procedure for the same period. The passive samplers are normally exposed in pairs, one for actual sampling and another for field blank. In this study field blanks were not collected at each site, but collected at four key representative sites. No duplicate samples were collected in this study. However, Ogawa passive samplers have been proven to yield reproducible results in the past (Roadman et al., 2003;Meng et al., 2010Meng et al., , 2011. As particulate matter loading is quite high in ambient air in the valley, the samplers were replaced every week before the sampling pads start getting clogged with deposition of particles, which could reduce diffusion of gases and hence introduce measurement artifacts.

Sampling during 2013
During the Figure 14). Further classification was implemented based on an approximate distance of a given site from the 28 km circle-like Ring Road surrounding Kathmandu metropolitan city and the northern part of Lalitpur district, including Lalitpur sub-metropolitan city. The sites located within this Ring Road are defined as urban sites. The immediate neighborhood setting of the site and distance from the urban core (i.e. area enclosed by Kathmandu's Ring Road) were also taken into account while classifying the sites.
During the same sampling period, ozone data were collected at a stationary site using online ozone monitor (Model-400E, Teledyne Technologies, Inc., USA) collocated with a passive sampler at one of the sampling sites, Bode, that served as the supersite for SusKat-ABC campaign (Naja et. al, 2015). Among the 15 sampling sites, 5 were classified as urban, 4 as suburban, and 6 as rural sites (Table 2). NO was calculated as the difference between NOx and NO 2.

Meteorological parameters
Meteorological parameters such as temperature, relative humidity, amount of rainfall, wind speed and direction were monitored at 6 locations that overlapped with sampling sites (Bode, Pakanajol, Nagarkot, Naikhandi and Bhimdhunga) and also from publically available meteorological data from Tribhuvan International Airport.

Laboratory analysis
After the completion of sampling, filters were carefully removed from Ogawa samplers and stored in air-tight bottles. Samplers were stored in Nepal for approximately 5 months prior to being shipped to our laboratory for chemical analysis. The samples, including field blanks, were digested in ultrapure water. Extracted NOx and NO 2 samples were analyzed using a spectrophotometer (SpectraMax-M2, USA), while O 3 , SO 2 and NH 3 extracts were analyzed using Ion Chromatography (Dionex-ICS-1100, USA).
During analysis, internal standards and laboratory blanks were analyzed in between every ten samples for quality control; field blanks were treated as unknowns. Method Detection Limits (MDL) were calculated as three times the standard deviation of a lab blank concentration (µg/ml) for each measured compound. Laboratory blanks and a selected standard were analyzed in between every 10 th sample for all measured species during their respective laboratory analysis techniques. For standards and laboratory blanks liquid concentrations (in µg/ml) were propagated to µg/m 3 by assuming typical measurement conditions, 10000 minute sampling time, 20 degrees Celsius, 40% Relative Humidity and, associated conversion coefficients.
Using the equations in Ogawa sampling protocols shown in section 2.  For ammonia, the lowest method detection limit has been reported to be 3.7 µg/m3 for a 1-day deployment and 0.26 µg/m 3 for a 14-day sampling duration (Roadman et al., 2003), while the upper detection limit (293 µg/m 3 ) have been found in another study for a 14-day sampling duration (Puchalski et al., 2011).
I was able to assess the detection limits provided by the manufacturer (Table 3) and also by the studies mentioned in the above paragraph for ammonia based on the calibration information I generated (Section 2.1.2) For example, the upper limit can be estimated from the highest laboratory standard while lower detection limit were assessed by computing three times the standard deviation of each pollutant detected in laboratory blanks and then propagating the liquid concentrations (in µg/ml) to atmospheric concentration in µg/m 3 .

Concentration conversion equations
Gas specific conversion equations: Nitrite solution) was run on each day samples were analyzed to check for instrument precision.
And the relative standard deviation (RSD) of this internal standard was found to be 6.2% among 4 duplicate runs of this particular standard.
The average Nitrite concentrations detected in laboratory blanks was 0.62 ± 0.42 µg/m 3 for NOx with values ranging from 0.07 -1.08 µg/m 3 . Each blank value (in µg/m3) was subtracted from the respective 10 unknown samples analyzed before the blank sample in a given analysis date. This blank correction procedure was applied for each analyzed gaseous specious.
The MDL for NOx, calculated as three times the standard deviation of a lab blank concentration for the measured compound, was 2.1 µg/m 3 . This was slightly higher compared to the NOx LOD from the manufacturer (1.01 µg/m 3 ).   The middlemost standard (i.e. 1.6 µg/ml nitrate solution) was analyzed in between every 10 th unknown sample on each day that samples were analyzed to check for instrument precision.
And the relative standard deviation (RSD) of this internal standard was found to be 1.58% among 16 duplicate runs of this particular standard.
The internal standard concentrations obtained from IC analysis were propagated to µg/m 3 by assuming typical measurement conditions. The resulting ranges of concentrations were  In this study field blanks were not collected at each site, other than at five key representative sites (namely: Bhimdhunga, Bode, Budhanilkantha, Nagarkot and Naikhandi).
The average field blank concentrations of pollutants at each available site are presented in the table below (Table 4). For most measured elements, field blank concentrations were consistent, but slightly higher than laboratory blanks. An exception to this was ozone, where field blanks were significantly higher. The limited field blanks shown here were not subtracted in calculation since concentration of measured pollutants was highly variable across and within sites and thus doing so would increases uncertainty. Therefore, only lab blanks are included in concentration calculations since ten out of fifteen sites did not have field blanks.

Variability across sites
Wide variability of trace gases was observed in the Kathmandu Valley across time, location, and measured species during the sampling periods. The weekly mean concentrations for each gaseous pollutant measured during 2013 sampling period is presented in Table 5
Concentrations of these gases were highly variable with observed weekly concentrations in the valley ranging from 12.4 -0.4, 46.0 -0.2, and 46.7 -4.1 µg/m 3 for NO, NO 2 , and NOx, respectively. The measurements were statistically different among these three site types (ANOVA test, p<0.05). However, intergroup comparison using an ANOVA post-hoc analysis (with Bonferroni correction) showed a statistically significant difference between urban and suburban, and urban and rural site types, but not between suburban and rural site types. Further classification of site types based on distance (in km) from main core ring road with high traffic activity is shown in Figure 16. Sampling locations within 2 km of Ring Road showed the highest levels of NOx (ANOVA test, p<0.05). Figure 16: 2013 mean concentrations (µg/m 3 ) of NO 2 , NOx and NO categorized by distance (in km) from main core ring road with high traffic activity. Error bars = standard deviation Note: Sites within 2 km from core ring road include: Maharajgunj, Suryabinayak, Mangal Bazaar, Indrachowk, Kirtipur, Bhaisepati and Bode. Sites located between 2 km and 6 km include: Budhanilkantha, Bhimdhunga, Naikhandi, Tinpiple and Lubhu. Sites located 6 km or more include: Sankhu, Nagarkot and Nala Pass.

Sulfur Dioxide
SO 2 concentrations during 2013 varied from site to site as shown in Table 5. The weekly mean SO 2 concentrations at urban sites (16.6 ± 12.9 µg/m 3 ) were higher than suburban (9.3 ± 2.1 µg/m 3 ) and rural sites (6.9 ± 2.5 µg/m 3 ) ( Figure 15). However, the spatial variation of SO 2 did not show statistically significant differences among site categories when an ANOVA test was performed.
The average SO 2 concentrations between Bode, Suryabinayak and Lubhu were in the range of 39.2-12.0 µg/m 3 with the highest weekly average SO 2 concentration observed at Bode (39.2 ± 21.6 µg/m 3 ) and Suryabinayak (15.8 ± 7.4 µg/m 3 ), both urban locations, and somewhat lower levels in Lubhu (12.0 ± 5.1 µg/m 3 ), a suburban site. These three sites are located within 3 km of at least 10 brick factories (Figure 17). In contrast, sampling sites located with similar proximity but to fewer brick kiln factories are observed with lower SO 2 levels. Such sites included Kirtipur and Naikhandi where highest weekly averages levels reached 9.4 ± 7.2 µg/m 3 and 11.7 ± 7.7 µg/m 3 , respectively. Sub categorizing these site types based on proximity to brick kilns showed no statistical significance. The lowest average concentration of SO 2 was measured at Bhimdhunga with an overall average of 4.7 ± 1.6 µg/m 3 .

Figure 17: 2013 mean concentrations (µg/m 3 ) of SO 2 categorized by district and sampling site location with respect to distance from brick kiln factories. Error bars = standard deviation
Note: Within 3Km (<10 brick kilns): Sites included in this category are: Naikhandi, Bhaisepati and Kirtipur. These sites are located within 3 km of less than 10 sparsely located Brick Kiln factories. Within 3Km (10+ brick kilns): Sites located within 3 Km of included in Bhaktapur district include Lubhu from Lalitpur district. Sites included in this classification are: Bode, Suryabinayak and Lubhu. These sites are located within 3 km of more than 10 densely located Brick Kiln factories. Within 3Km (No brick kilns): Sites included in this category are: Sankhu, Nagarkot, Tinpiple and Bhimdhunga. No brick kiln factories are present within 3 km of these remote sites.

Ozone
The weekly mean concentrations of ground level ozone at rural sites (110.7 ± 17.0 µg/m 3 ) were higher than suburban (95.9 ± 3.7 µg/m 3 ) and urban sites (84.9 ± 10.2 µg/m 3 ) as shown in Figure 15. Variations across site type are significantly different (p=0.01) from one another as indicated by ANOVA test, however, post-hoc analysis showed that significance was

Measurements during 2013
Week to week variation in the valley during 2013 is presented in Table 6. In rural sites, NO 2 concentration variation between the first two sampling weeks (i.e. March 23 rd to April 6 th , 2013) and the last two sampling weeks (i.e. May 4 th to May 8 th , 2013) was found to be significant (ANOVA test, p<0.05). NO 2 levels in urban and suburban regions showed higher concentration in May than March, although not statistically significant. NOx and NO levels showed higher concentrations in March than the late periods of the pre-monsoon season, and the temporal variations showed significance for NO at all sites (ANOVA: p=0.001, p=0.003, p=0.01 at urban, suburban and rural, respectively). NO 2 forms quickly from on-and off-road vehicular emissions and is a good indicator for the larger group of oxides of nitrogen (Lippmann et al., 2003). Since automobiles are the major contributors of oxides of nitrogen to ambient air pollution (Rakowska et al., 2014) (Shrestha et al., 2013). This fact provides the basis to further classify and investigate the spatial association shown in Figure 16. Sampling locations within 2 km of Ring Road have the statistically significant highest levels of NOx (ANOVA test, p<0.05). Although sampling locations within 2 km of Ring Road had the highest levels compared to sites between 2-6km and 6km, post-hoc analysis did not show meaningful difference amongst the other groups. NO 2 is emitted directly from combustion processes and also results from oxidation of NO (Yao et al., 2005). At the urban sites, slightly elevated levels of NO 2 /NOx ratio were found (80.2 ± 8.7%) compared to that of suburban sites (78.8 ± 4.3%) and rural sites (76.2 ± 8.0%). Ratios reaching as high as 95 % and 96 % in Indrachowk and Suryabinayak, respectively, were measured while the lowest ratios were observed in Nagarkot (37%) and Lubhu (21%). The increased NO 2 /NOx ratios at urban sites indicate that there is a significant contribution of primary NO 2 from on-road vehicle sources to ambient air pollution. Previous studies have found that, under normal driving conditions, the reported primary vehicular NO 2 /NOx ratio of gasoline engines varied from 2-5% to approximately 30% (Yao et al., 2005). This volume ratio has been found to be substantially higher for diesel engines. Tang et al. (2004) report volume ratios as high as 50% for diesel buses.
It is interesting to note the observed NO 2 /NOX ratios in the Kathmandu Valley which were higher than what many previous studies have reported. Carslaw (2005)  combination of these effects that lead to high NO 2 /NOx ratios, though week-long passive samples as used in this work may not be directly comparable to previous tunnel studies with high time resolution and thus are unlikely to provide any better insight or evidence necessary to test these hypotheses. Thus, more work in this area is needed to explain this observation.
In the valley, alternative fuel sources such as batteries and LPG (Liquefied Petroleum Gas) used in light duty vehicles such as vans and microbuses make up only a small share of transport fuel consumption, while gasoline and diesel-powered vehicles dominate the valley (Shrestha et al., 2013).
High NO levels and low O 3 concentrations observed at urban sites during the winter season suggest that ground level ozone most likely undergoes rapid NOx titration in urban areas.
Studies have found that places in the valley located in the immediate vicinity of very large NO emission, such as high traffic intensity areas, experience O 3 concentration depression through reaction with NO (Pudasainee et al., 2006(Pudasainee et al., , 2010. Thus, it is quite plausible that high intensity traffic sources explain the observed higher NO 2 near urban core Ring Road. In contrast, vehicle emission might be of lesser importance in rural sites such as Nagarkot, Naikhandi and Bhimdhunga, however, rural sites located downwind from Kathmandu city and northeast section of the valley (i.e., Sankhu and Nala Pass) show surprisingly higher NO 2 levels. Prevailing southwesterly winds were dominant during the winter sampling period (Figure 19) indicating that these compounds are advected from the urban areas to downwind sites, even if there aren't many local sources at downwind locations. Additional significant sources at these rural sites may be NOx emission from brick kiln factories and anthropogenic combustion from suburban residential sites located within 1 km of these rural sites.

Sulfur Dioxide
Coal combustion in brick factories are likely to contribute to elevated ambient levels of SO 2 . More than 100 brick kilns throughout the valley produce over 350 million bricks per year, and they are significant emission sources of SO 2 and PM in the Kathmandu Valley, contributing to over 60% of total SO 2 and particulate matter emissions (Joshi andDudani, 2008, Pariyar et al., 2013). Moreover, 74% of total anthropogenic SO 2 emissions in the valley result from combustion in the manufacturing industry that is mainly comprised of brick factories and other petroleum products . On the other hand, commercial and residential sectors accounts for just 16% of total anthropogenic SO 2 emission in the valley , indicating that SO 2 sources are mainly point sources such as brick kilns.
The average SO 2 concentrations between Bode, Suryabinayak and Lubhu were in the range of 39.2-12.0 µg/m 3 with the highest weekly average SO 2 concentration observed at Bode (39.2 ± 21.6 µg/m 3 ) and Suryabinayak (15.8 ± 7.4 µg/m 3 ), both urban locations, and somewhat lower levels in Lubhu (12.0 ± 5.1 µg/m 3 ), a suburban site. In addition to the increased presence of mobile and stationary generators that use low-quality (i.e. high sulfur content) diesel fuel in urban/suburban sites, these three sites are located within 3 km of at least 10 brick factories ( Figure 17), which typically use high quantities of low grade coal in brick production (Raut, 2006, Pariyar et al., 2013. In contrast, sampling sites located with similar proximity but to fewer brick kiln factories are observed with lower SO 2 levels. Such sites included Kirtipur and Naikhandi where highest weekly averages levels reached 9.4 ± 7.2 µg/m 3 and 11.7 ± 7.7 µg/m 3 , respectively. The lowest average concentration of SO 2 was measured at Bhimdhunga, a mountain pass on the western rim of the valley, with an overall average of 4.7 ± 1.6 µg/m 3 . Bhimdhunga is a rural location with no nearby brick kilns and other major SO 2 sources. Apart from mandatory gravity settling chambers on brick kiln factories, no other pollution control devices are in operation on 111 known brick kilns, 89 known stone crushing factories and 70 industries with boilers (Dhimal, 2009). Brick kiln facilities are concentrated in the southern and southeastern regions of the valley (Joshi and Dudani, 2008;Raut, 2006), this suggests a plausible source of SO 2 observed in high concentrations at Bode, Suryabinayak and Lubhu, all of which are located in the south and southeastern region of the valley and close to the brick kilns. High levels of SO 2 in Naikhandi (11.7 ± 7.7 µg/m 3 ), one of the southern locations in the valley, might be attributed to 6 brick kiln factories located within 3 km of this sampling site.
Other studies in Kathmandu have also found elevated SO 2 levels (3.3 -23.4 µg/m 3 ) in southeastern region of the valley where over several brick kilns are located (Byanju et al., 2012;Pradhan et al., 2012). Much higher SO 2 levels were also measured in 2001 dry season (36.7-78.6 µg/m 3 ) at sites mainly surrounded by brick kilns .

Ammonia
NH 3 concentrations varied from site to site as shown in Table 5; however, the variation was not significantly different across site categories. This might be attributable to fewer emissions of NH 3 from industries such as coke and ammonia production factories , compared to SO 2 and NO 2 . NH 3 plays a key role in the formation and neutralization of acidic atmospheric pollutants such as nitrates and sulfates (Sharma et al., 2007). About 47% of anthropogenic NH 3 emission in the valley comes from agriculture residues, soil emission, burning and manure management . In addition, the waste sector, which includes municipal waste and incineration, contributes 36% to the anthropogenic NH 3 emissions in the valley .
High NH 3 concentrations in Suryabinayak are likely attributed to fertilizer use and ammonia emissions from animal manure and chemical fertilizer used in nearby agricultural fields. In urban locations like Indrachowk these agricultural sources are not likely important, and it is possible this NH 3 is linked to municipal waste disposal and incineration. Dahal et al., (2011) have indicated increasing levels of municipal waste contamination in rivers throughout Kathmandu. Indrachowk, for instance, is located approximately half a kilometer away from highly polluted Bishumati River that cuts through the urban core. Other urban locations, such as Budhanilkantha and Maharajgunj, which are away from such rivers had lower NH 3 concentrations, and tended to have more typical of regional background levels indicated by the rural sample locations. These levels were typically around 19-22 µg/m 3 on average. Generally, regions located downstream of major rivers in the valley had elevated NH 3 levels (Figure 18), although one-tailed t-test showed no significance (p=0.09) compared to upstream sites.
Additional sources of ammonia in the valley include emissions from ammonia-based refrigeration and ammonia-based solvent manufacturing factories, and naturally from microorganisms involved in decaying animal matter and also in sewage treatment , including highly polluted rivers themselves, though these emission sources are poorly documented. Extensive sampling is needed near these polluted rivers in Kathmandu valley in order to observe a clear association between NH 3 levels and contaminated rivers. Since this was not planned in the SuSKat study design, our assessment of NH 3 near polluted rivers is based on fairly limited data.

Ozone
The weekly mean concentrations of ground level ozone at rural sites (110.7 ± 17.0 µg/m 3 ) were higher than suburban (95.9 ± 3.7 µg/m 3 ) and urban sites (84.9 ± 10.2 µg/m 3 ) as shown in Figure 15. It is not surprising to see ozone at higher levels at sites downwind of urban outflows. In Kathmandu Valley, there are many unregulated sources of ozone precursor gases (Sapkota and Dhaubhadal, 2002) including a high number of vehicles (Gurung and Bell, 2012), and fossil fuel and biomass combustion in manufacturing industries such as brick factories (Raut, 2006). NOx and Volatile Organic Carbon (VOC) emissions from such sources in the valley, as indicated in our measurements and past studies in the valley (e.g., Pudasainee et al., 2006) are likely sufficient to support ozone formation. The proportional contribution of NOx and several VOCs to O 3 formation in the valley requires further investigation.
Higher altitude locations also had elevated O 3 concentrations. For instance, Nagarkot and Bhimdhunga, the two most elevated sampling sites located at an elevation of 1895 m and 1530 m above sea level had highest mean (8 week) concentration of 137.5 ± 19.1 µg/m 3 and 117.0 ± 23.4 µg/m 3 , respectively. One possible explanation could be elevated levels due to in-situ or in-transport conversion of its precursors such as NOx and VOCs. Other studies have also found elevated ozone levels at high altitude rural sites, for example, in Spain (Sánchez et al., 2005) and in India (Naja et al., 2003).

Measurements during 2013
Increased NO level in urban areas indicates the high variability of this gas with respect to traffic intensity and meteorological conditions such as amount of solar radiation compared to suburban and rural areas. One possible explanation for the higher NOx levels in March could be the seasonal component of oxides of nitrogen species, where drier and colder conditions contribute to the presence of many other reactive nitrogen oxides, other than NO 2 . The mixing ratio of each component of nitrogen oxide species is observed being the highest during the driest seasons in Quebec, Canada, due to seasonal cycle partitioning of NO y (total odd nitrogen) that includes HNO 3 , NO 3 and N 2 O 5 (Hayden, 2003).
Another interesting observation is that NH 3 , O 3 and SO 2 concentrations in the valley were higher during the relatively dry first half of the sampling duration, particularly during the third week of sampling (April 6 th -13 th , 2013) than rest of the sampling period. Winter and sunny days during late March and early April possibly contributed to factors leading to these highest levels, May 18, 2013. Comparison of these two sampling methods indicated that passive sample measurements of ozone were likely to be biased low, on the order of a factor of 20-30%, possibly due to deposition of particles on the sampling pad that might have reduced the diffusion process.
However, passive samplers have been used extensively to monitor ambient O 3 in Europe (Bytnerowicz et al., 2002a) and North America (Bytnerowicz et al., 2002b). Nonetheless, this has important public health relevance as O 3 levels higher than 100 µg/m 3 are associated with 1-2% increase in daily mortality (WHO, 2005). The highest monthly peaks were recorded in April, and the lowest in late March and May as shown in Figure 20. Pudasainee and colleagues (2006) found that enhanced ozone concentrations were due in part to long sunshine hours during the spring season (i.e. March to May) in the valley, prior to the more frequent cloud cover prevalent during the rainy season. Ozone measurements collected in 2004 (Pudasainee et al., 2006) at Pulchowk (an urban site 7 km west of Bode) reported that the hourly levels of ozone during the same spring season peaked at a maximum of 120 µg/m 3 . Our findings, on average, were observed to be somewhat higher than previous measurements with hourly maximums peaking up to 200 µg/m 3 ( Figure 20). This suggests that ozone levels in the valley have increased over time over the past decade. Other studies in neighboring countries have also found high levels of ozone in Delhi, India (Jain et al., 2005) and increased background levels of ground ozone by 0.5% per year from 1950s to 1990 in Ahmedabad, India (Naja and Lal, 1996) indicating that this may be a regionally important issue. where ozone was observed at lower levels within the urban regions and at high concentrations, often exceeding the WHO recommended guidelines, downwind of these urban centers.
In the context of public health and exposure to these pollutants, it is known that exposure to these gasses has been associated with a variety of adverse health effects, especially when levels exceed a certain threshold. Our measurements from 15 sites across the valley clearly indicate that O 3 , which is toxic to both human and plants, is an important pollutant of concern that demands urgent attention. In Kathmandu valley, amongst all of the measured pollutants, we have observed frequent threshold exceedance, particularly for ozone. This has important public health relevance for the people residing in Kathmandu valley as exposure to ozone levels higher than 100 µg/m3 (i.e. WHO guideline) have been shown to cause several health outcomes such as, rise in daily mortality and increased hospital admissions rates due to illnesses such as pneumonia and COPD. Therefore, due to the health detriments attributed to these gaseous pollutants, ozone predominantly, this pollutant is an important pollutant of concern that requires immediate attention and control through mitigation of its precursors such as nitrogen oxide species in the valley.
This work provides a broad analysis of anthropogenic gases by incorporating multiple sites over the wide range of the Kathmandu Valley, a region known for its notorious particulate matter pollution. It is still the main pollutants that pose serious health risks to 3.5 million residents of the valley, and thus requires priority mitigation actions. These results could be useful to aid urban planning and pollution mitigation efforts that will encourage cleaner, healthier, more sustainable and livable cities across the Kathmandu Valley.

Limitations leading to alternative/ complementary sampling techniques
Globally, more than 3 million premature deaths occur annually due to outdoor air pollution; more than 80% occurring in the developing world (WHO, 2012). Because most of our understanding of this morbidity arises from studies in the developed world, dose-response relationships might not be applicable for exposures in highly polluted areas, such as Nepal.
Hence, from a public health standpoint, increased measurements of air pollutants in outdoor environments is critical for assessing exposure to air pollution and the potential health effects in low income countries.
Due to their simplicity, low cost and capabilities of providing similar performance to expensive active samplers in terms of sensitivity and reproducibility, passive samplers have been widely used as an effective alternate for conventional active samplers in exposure and health effects studies, especially in developing countries. However, one of the major limitations to passive samplers is a long sampling time (often integrated over 7 days at a minimum) is usually required in order to get enough mass for detection. Although these easily accessible and inexpensive hardware and software microprocessors toolkits were originally created for applications in actuator control functions such as in simple robotics, their discovery has allowed for professionals along with citizen scientists to gain ease of access for operating low-priced environmental sensors. A good example here can be the use of semiconductor metal oxide (SMO) gas sensors that have been widely used in the auto industry for monitoring vehicle cabin air quality (Galatsis and Wlodarski, 2006). In recent times, these SMO gas sensors have found other applications in measuring traces gases such as carbon monoxide and NOx (Fine et al., 2006), and ozone (Gerboles and Buzica, 2009) in the ambient atmosphere with low-priced data processing platforms able to process, display, and store sensor signals within seconds.
The applicability of such inexpensive environmental sensors and processing platforms is exciting. An Arduino, for instance, is equipped with simple programming language, open source software, digital and analog pins which can be interfaced to a variety of expansion boards and circuitry embedded in, for instance, an electrochemical or a metal oxide gas sensor or an optical particle counter. This can be assembled and coded by anyone with internet access and required no significant background in circuitry or measurements.
Since the turn of the new millennia, low-cost air sensors have continuously entering the global market and are commercially available in a wide variety of design and capabilities. Recent studies have shown there is a potential in exploring low-cost small sensors use in field studies.

A.IV: Discussion (Variability across sites)
Urban sites during the winter season had very high concentrations of oxides of nitrogen, with weekly averages frequently exceeding the 24-hour NAAQS guideline for NO 2 (80 µg/m 3 ).
Though not directly comparable, more than 30% of total 2014 winter season weekly measurements exceeded the 24-hour NAAQS level. This is not surprising as these are roadside measurements taken from main traffic intersections with heavy traffic movement in the valley.
Large primary NO 2 concentrations were observed from on-road vehicle sources with NO 2 /NOx ratios reaching as high as 58% at urban on-road sites in both seasons (43 ± 9.0 % during winter season, and 46 ± 17.0 % during monsoon season). Lower O 3 levels measured in 2014 compared to 2013 levels, particularly in urban areas, attribute to high NO concentrations in 2014 suggesting that ground level ozone was scavenged by NO titration. Studies have found that sampling sites located in the immediate vicinity of very large NO emission, such as high traffic density areas, experience high removal of ozone through reaction with NO (Derwent et al., 2003;Pudasainee et al., 2010).
SO 2 levels remained low throughout the dry season (Table 8, and Figure 22). Although a large fraction of the data was below the detection limit, higher average SO 2 levels were observed during the rainy season (25.4 ± 11 µg/m 3 ) compared to winter season (6.7 ± 0.4 µg/m 3 ) at urban sites. As discussed earlier, the major source of SO 2 in suburban and rural locations is brick production, an industry which does not normally operate in rainy season and hence the SO 2 concentrations were expected to be lower in rainy season. Also, frequent rain events wash away SO 2 from the atmosphere through acidic rain formation. Incomplete site type and missing data limits our spatial variability interpretation of the SO 2 concentrations observed during the winter season alone. In contrast, spatial variability is observed during the monsoon season ( Figure 23).
Mean SO 2 levels were lower in urban areas (0.7 ± 0.7 µg/m 3 ) compared to suburban (4.3 µg/m 3 ) and rural (5.0 µg/m 3 ) sites. O 3 levels in the monsoon season (14.0 ± 1.1 µg/m 3 ) remained similar among all site types and also were comparable to winter season (14.0 ± 9.6 µg/m 3 ). Moreover, mean concentrations of NO 2 at urban sites (21.4 ± 8.7 µg/m 3 ) in the monsoon season were higher than suburban sites (NO 2 4.9 ± 1.0 µg/m 3 ) while mean concentrations of NOx and NO at urban sites (NOx 23.6 ± 11.1 µg/m 3 , NO 4.4 ± 0.4 µg/m 3 ) were lower than suburban sites (NOx 26.6 ± 28.8 µg/m 3 , NO 21.7 ± 29.8 µg/m 3 ) during the same monsoon season. Large NO deviation observed in (Figure 22) indicates that one site in particular, Kirtipur (Table 8), was responsible for deriving the average levels of NO and NOx to much higher levels. It is possible that highly localized sources emitting high levels of NO existed near sampling location in Kirtipur-perhaps a momentary spike in traffic activity at Tribhuvan University-during the time of measurement.
As expected, lowest levels of NOx and NO (6.2 ± 1.1 µg/m 3 and 1.0 µg/m 3 , respectively) were observed in rural sites; although rural NO levels were measured at one site only. NO 2 concentrations at rural sites (6.1 ± 0.9 µg/m 3 ) were slightly higher than suburban sites (4.9 ± 1.0 µg/m 3 ), however, this can be explained either by NO 2 long range transports (Zien et al., 2014) or a highly localized source during the measurement week. The effect could have also been observed due to limited number of sampling sites used in the site type stratification.

A.V: Discussion (Temporal variations)
The overall average concentrations of trace gases for 2014 winter and monsoon seasons are shown in Figure 23. From the 2014 measurements, it is interesting to note that monsoonal onroad concentrations were no lower (with the exception of SO 2 ) than compared to winter season trace gas levels (Figure 23), where it was expected concentrations would be higher. Pollutant levels were of the same order of magnitude in two cases (i.e. NO and NO 2 ) and were higher in other two (i.e. O 3 and SO 2 ) during the monsoon season of 2014 ( Figure 23). Similar NO 2 levels in the two seasons, likely because the focus of this study was on-road emissions in busy roadways. Our findings of NO 2 levels (103.9 ± 15.7 µg/m 3 and 102 ± 51.2 µg/m 3 during, dry and wet season, respectively) are at least a factor of two higher than the urban on road NO 2 levels reported in 2008 (18.9-52.6 µg/m 3 and 12.4-33.6 µg/m 3 during dry and rainy season, respectively) (Byanju et al., 2012).
Although not particularly evident from 2014 sampling period due to limited samples size, ± 11 µg/m 3 and O 3 16.6 ± 6.5 µg/m 3 ) compared to the winter season (SO 2 6.7 ± 0.4 µg/m 3 and O 3 14 ± 9.6 µg/m 3 ). This is contrary to the previous study (Byanju et al., 2012) that reported higher levels of SO 2 during the winter season. it should be noted that fewer samples of SO 2 are available in this study because 50% of collected on-road samples were below the detection limit (Table 8).
Comparing SO 2 levels at Jawalakhel, where both monsoon and winter levels are available (Table   8), it is possible that an air pollution episode might have occurred during the second week of August 2014 causing levels of pollutants to be elevated. At Jawalakhel, NO levels during monsoon season were below the detection limit possibly because of reaction with high O 3 , suggesting high ozone levels (Table 8). In addition, lowered O 3 levels during the winter season could be partly explained by the presence of higher NO levels in the winter season compared to the monsoon season; leading to enhanced O 3 titration by NO. These five on-road urban sites are in the immediate vicinity of very large NO emissions from high traffic intensity, and such levels are pronounced during the winter season where concurrent ozone concentration depression is often observed (Pudasainee et al., 2006(Pudasainee et al., , 2010. Thus, it is likely that high intensity traffic sources explain the temporal variations observed among NOx species and O 3 levels.