Airborne particulate matter monitoring in Kenya using calibrated low cost sensors

8 East African countries face an increasing threat from poor air quality, stemming from rapid 9 urbanisation, population growth and a steep rise in fuel use and motorization rates. With few air 10 quality monitoring systems available, this study provides the much needed high temporal resolution 11 data to investigate the concentrations of particulate matter (PM) air pollution in Kenya. Calibrated 12 low cost optical particle counters (OPCs) were deployed in Kenya in three locations: two in the 13 capital of Nairobi and one in a rural location in the outskirts of Nanyuki, which is upwind of Nairobi. 14 The two Nairobi sites consist of an urban background site and a roadside site. The instruments were 15 composed of an AlphaSense OPC-N2 optical particle counter (OPC) ran with a raspberry pi low cost 16 microcomputer, packaged in a weather proof box. Measurements were conducted over a two- 17 month period (February – March 2017) with an intensive study period when all measurements were 18 active at all sites lasting two weeks. When collocated, the three OPC-N2 instruments demonstrated 19 good inter-instrument precision with a coefficient of variance of 8.8±2.0% in the PM 2.5 fraction. The 20 low cost sensors had an absolute PM mass concentration calibration using a collocated gravimetric 21 measurement at the urban background site in Nairobi. 22 The mean daily PM 1 mass concentration measured at the urban roadside, urban background and 23 rural background sites were 23.9, 16.1, 8.8 µg m -3 . The mean daily PM 2.5 mass concentration 24 measured at the urban roadside, urban background and rural background sites were 36.6, 24.8, 13.0 25 µg m -3 . The mean daily PM 10 mass concentration measured at the urban roadside, urban background 26 and rural background sites were 93.7, 53.0, 19.5 µg m -3 . The urban measurements in Nairobi showed 27 that particulate matter concentrations regularly exceed WHO guidelines in both the PM 10 and PM 2.5 28 size ranges. Following a ‘Lenschow’ type approach we can estimate the urban and roadside 29 increments that are applicable to Nairobi. Median urban and roadside increments are 33.1 and 43.3 30 µg m -3 for PM 10 , respectively, the median urban and roadside increments are 7.1 and 18.3 µg m -3 for 31 PM 2.5 , respectively, and the median urban and roadside increments are 4.7 and 12.6 µg m -3 for PM 1 , 1 respectively. These increments highlight the importance of both the urban and roadside increments 2 to urban air pollution in Nairobi. 3 A clear diurnal behaviour in PM mass concentration was observed at both urban sites, which peaks 4 during the morning and evening Nairobi rush hours; this was consistent with the high measured 5 roadside increment indicating that vehicular traffic is a dominant source of particulate matter in the 6 city, accounting for approximately 48.1, 47.5, and 57.2% of the total particulate matter loading in 7 the PM 10 , PM 2.5 and PM 1 size ranges, respectively. Collocated meteorological measurements at the 8 urban sites were collected, allowing for an understanding of the location of major sources of 9 particulate matter at the two sites. The potential problems of using low cost sensors for PM 10 measurement without gravimetric calibration available at all sites are discussed. 11 This study shows that calibrated low cost sensors can be used successfully to measure air pollution 12 in cities like Nairobi. It demonstrates that low cost sensors could be used to create an affordable and 13 reliable network to monitor air quality in cities. m 3 h -1 . PM 10 is therefore the sum of the two size fractions (PM 2.5 + PM 10-2.5 ). The chosen sample day 27 was rain free and had similar temperature and RH profiles compared to the rest of the OPC sampling 28 campaign. The filters were weighed using a mass balance before and after particulate matter 29 collection. The observed 24 h average mass concentrations of PM 2.5 and PM 10 from the impactor 30 were 27.6 ± 6.8 and 51.8 ± 10.3 µg m -3 , respectively, while those recorded from the OPC 16.9 and 31 30.6 µg m -3 , respectively. The uncertainty in gravimetric concentrations was estimated from the 32 instrument (10%), sampling (7%) and weighing (25%) errors and that of the OPC data was the 17 average PM mass concentration observed in this study suggests that that the WHO 1 recommendations for annual PM 2.5 and PM 10 are likely exceeded at both the urban background and 2 urban roadside locations. For the urban background site, the measured average PM 2.5 and PM 10 3 mass concentrations exceed the annual WHO recommendations by factors of 2.5 and 2.7, 4 respectively. Whereas for the urban roadside site they exceed recommendations by 3.7 and 4.7, 5 respectively. These significant exceedances for both the urban roadside and urban background sites 6 suggests that most of Nairobi’s population will be subjected to outdoor air pollution far in excess of 7 the WHO recommendations for annual exposure. Figure 3 provides the box and whisker plots for the 8 hourly averaged PM 2.5 and PM 10 data for the three measurement sites, highlighting the proportion of 9 the days which exceed the WHO annual and daily recommendations. This paper used calibrated OPC-N2 devices to measure PM concentrations in Nairobi, Kenya in the 4 size fractions PM 1 , PM 2.5 and PM 10 . The data required calibration using an established gravimetric 5 approach to PM measurement. The need for calibration by trained scientists significantly increases the costs associated with low cost monitoring and this cost needs to be factored in when assessing 7 options for air quality monitoring.


p r o v i d e d t h a t
• t h e a u t h o r s, ti tl e a n d full bi blio g r a p hi c d e t ails of t h e it e m a r e ci t e d cl e a rly w h e n a n y

You m a y n o t
• s ell a n y p a r t of a n it e m • r ef e r t o a ny p a r t of a n it e m wi t h o u t ci t a tio n • a m e n d a ny it e m o r c o n t ex t u ali s e it in a w a y t h a t will i m p u g n t h e c r e a t o r 's r e p u t a tio n • r e m ov e o r al t e r t h e c o py ri g h t s t a t e m e n t o n a n it e m .
Th e full p olicy c a n b e fou n d h e r e . Alt e r n a tiv ely c o n t a c t t h e U niv e r si ty of C u m b ri a R e p o si t o ry E di t o r by e m aili n g in si g h t @ c u m b ri a. a c. u k .  Airborne particulate matter (PM) air pollution is a major environmental risk factor with well-28 documented short and long-term effects on human mortality and morbidity (Thurston et al., 2016). 29 It is known to affect asthma, chronic pulmonary disease (COPD), pulmonary fibrosis, cancer, type-2 30 diabetes, neurodegenerative diseases, obesity and other conditions (Ferranti et al., 2017). The size 31 of PM is correlated with their health impacts, with smaller particles typically having more significant 32 health implications (Meng et al., 2013). PM 1 , PM 2. 5  air for much longer than coarser particulate matter, as well as penetrating deeper into the lungs 6 leading to local pulmonary, systematic inflammation (Pateraki et al., 2014). Due to the smaller size, 7 PM 1 has a higher surface to mass ratio, containing a harmful amount of potentially toxic 8 anthropogenic constituents which could lead to health impacts such as respiratory disease, heart 9 disease and lung cancer (Trippetta et al., 2016). Many studies still focus on PM 10 and PM 2.5 even 10 though smaller particulates pose greater health impacts (Tsiouri et al., 2015). Beyond PM 1 , ultra-fine 11 particles (<100 nm) are of such a small size they can be translocated to the central nervous system 12 via the blood to brain barrier or the olfactory bulb. There are no air quality regulations of PM 1 or 13 ultra-fine particles due to the paucity of data either within environmental science or public health. 14 Worldwide, road traffic is a dominant source of urban PM accounting for 5-80% of PM mass, with 15 the precise amount being dependent upon several factors including time, location, and vehicle fleet, 16 as reviewed by Pant and Harrison (2013). Vehicle derived PM is directly associated with negative 17 health outcomes (Fan et  In many LMIC cities, urbanization, population, fuel use and motorization rates are all increasing 32 rapidly and increases in air pollution are associated with these trends (Mitlin and Satterhwaite,33 (UNEP, 2005). Nairobi is the capital city of Kenya and is showing these trends. In 3 particular, the city population has increased dramatically, since 1999 to 2015 it has risen by 83%, and 4 is projected to increase to 7.14 million by 2030 (Rajé et al., 2017). Similarly, motorization rates are 5 increasing, between 2008 and 2012, the number of motor-and auto-cycles in Kenya grew by 368% 6 with the number of overall registered vehicles increasing by 77% (Rajé et al., 2017). Considering this 7 extensive increase in the vehicle fleet, limited roadway infrastructure and high congestion within the 8 city, pollution hotspots are created leading to personal exposure levels much higher than that 9 encountered throughout the rest of the city (van Vliet and Kinney, 2007). 10 To be able to reduce air pollution, you first have to be able to measure it. Many LMIC countries have 11 insufficient monitoring networks through which to measure air quality. In particular, long term high 12 resolution data is required for such cities which are vulnerable to air pollution. Nairobi is in the 13 vanguard of air pollution measurements for Sub-Saharan Africa but lacks continuous long term 14 calibrated measurements of PM and other air pollutants (Petkova et al., 2013). A discussion of the 15 relevant measurements in Nairobi is given in the next section. One of the constraints to making 16 measurements is the high cost of research grade air quality monitoring equipment with appropriate 17 calibration and certification. Low cost sensors offer the potential for dramatically reducing 18 equipment costs by orders of magnitude, making the monitoring of air quality more accessible and 19 attainable in LMIC countries Rai et al., 2017). 20 In this paper, the use of low cost sensors for measurement of PM 1 , PM 2.5 and PM 10 in Nairobi is 21 detailed. We have previously assessed the same low cost sensors in the UK (Crilley et al., 2018). The 22 sensors are calibrated using a standardised gravimetric approach. PM is measured in three locations: 23 an urban roadside site, an urban background site and a rural background site. Comparison of 24 simultaneous measurements at the three sites allows for the estimation of an urban increment and 25 roadside increment in PM following a 'Lenschow' type approach (Lenschow et al., 2001). The 26 variation of measured PM with measured meteorological data is also discussed. Finally, we discuss 27 the implications of using low cost sensors in Nairobi and LMIC countries in general. 28

Previous PM measurements in Nairobi 29
In general, long term air quality monitoring in Sub-Saharan Africa (SSA) is rare. Additionally, the study did not calibrate the monitors, which leads to questions about absolute 19 concentrations and interference from other environmental dependencies (Lewis and Edwards, 20 2016). The collected data from the study appeared noisy, with the authors stating they could not 21 separate the signal from the noise without having access to an air quality measuring reference 22 instrument (they recorded peaks at over 1000 µg/m 3 ). Despite the limitations, it provides a useful 23 comparison to this calibrated study. from Bedford, UK). They reported a concentration range of 3 µg/m 3 to 53 µg/m 3 at the urban 1 background site, with an overall mean of 21 µg/m 3 which exceeds the annual WHO limit of 10 µg/m 3 2 by a factor of two. The average concentrations of PM 2.5 at both sites were found to be 21 ± 9.5 and 3 13 ± 7.3 µg/ m -3 , respectively. Chemical composition measurements of the filter samples allowed 4 source apportionment, via positive matrix factorization, to be carried out. The analysis suggested 5 that five major source factors contribute to Nairobi PM 2.5 : traffic, mineral dust, industry, combustion 6 and a mixed factor. The dominant source factors were mineral dust and traffic which accounted for 7 74% of the particle mass. This study clearly highlights that the PM concentration in Nairobi varies considerably over both time 25 and space, which has significant implications for human exposure, see discussion.  Nairobi's longstanding popularity as a travel destination, due to its safari and other holiday resorts, 16 the city also acts as East Africa's diplomatic, financial and communication capital (Rajé et al., 2017). 17 Its geographical location is at approximately 1.29° S and 36.82° E. The highest elevation point in the 18 city is at an altitude of 1663m above the ground. As discussed in the introduction, Nairobi is 19 undergoing rapid increases in population and motorization both of which will likely lead to greater 20 PM pollution in the absence of any efforts of mitigation against the pollution. Other significant 21 infrastructure projects such as major road building are currently being undertaken, which will also 22 likely lead to increased PM loadings. Within Nairobi, the two field sites represent an urban 23 background location and an urban roadside location. The other site, a rural background site is 24 located on the outskirts of Nanyuki, a town that is located at an approximate aerial distance of 147 25 km to the north (NNE) of Nairobi and 240 km by road. The sensor boxes were placed in locations free 26 from obstacles, at the three measurement sites, allowing for 360 degrees of air flow. The second collection site in Nairobi was at the fire station, which is located within the CBD in the 7 city. Unlike the American Wing site, the area around the Fire Station is characterized by high traffic 8 flow which includes common public transport vans, locally known by the name "Matatus". It is on an 9 urban street canyon, on a street where smoking diesel vans are frequent and is exposed to urban 10 heat Island effects. It is also in the neighbourhood of vertical dispersion measurement site of PM 2. 5  The OPC-N2 is a miniaturized OPC which has dimensions of 75×60×65 mm and weighs under 105 g. 4 The unit cost of an OPC-N2 is approximately 250 GBP or 25000 KeS, hence it is significantly cheaper 5 than reference optical particle counter instruments which cost approximately 30-50 times as much. 6 Reference grade gravimetric instruments can cost even more. The lower cost of the OPC-N2, 7 provided the opportunity for measurements at multiple sites simultaneously. It measures particles in 8 the reported size range of 0.38 to 17 µm across 16 size bins, with a maximum particle count of 9 10,000 per second. The particle number concentration is converted by on-board factory calibration 10 to PM concentrations according to European Standard EN481 (OPC-N2 manual). 11

12
The assumed density for all particle sizes is 1.65 g/cm 3 and no special weighting is placed on any 13 particular bin size. However, the manual for the OPC states "an additional weighting is applied on 14 units with Firmware 18 or higher to account for under counting at low particle sizes and the effect of 15 carbon particles in urban air so that the output matches collocated reference detectors." 16 17 The lower cut off for particle size observed by the OPC is 380 nm and hence a large proportion of all 18 particles are not observed by the OPC due to the particle number being dominated by the smallest 19 particle sizes (Seinfield and Sypyros, 2016). Ultrafine particles (particles of aerodynamic diameter 20 <100 nm) were therefore not measured. However, the interest of the study was particulate mass 21 which is dominated by particle sizes that were measured. The local meteorology for Nairobi was measured at the same location as the urban background site 12 using a Vaisala instrument (WXT510) with the following variables measured: wind speed, wind 13 direction, temperature, relative humidity, relative humidity, barometric pressure, and rainfall with 14 an instrument temporal resolution of five minutes. The meteorology measured parameters were in 15 good agreement with other local measurements such as those observed at Jomo Kenyatta 16 International Airport (JKIA), which is approximately at an aerial distance of 10 km. The proximity of 17 the meteorological station at the urban background site to the urban roadside makes the 18 meteorological data appropriate for both sites. The data was collected at the urban background site 19 from the 2 nd of February to the 6 th of April 2017, covering the duration of the PM measurements. 20

OPC-N2 gravimetric mass calibration 21
The OPC-N2 mass concentrations were calibrated using gravimetric measurements of PM 2. 5  factors are likely due to differences in average particle densities observed in Kenya compared to that 5 observed in the UK, and also the typical RH measured in Nairobi compared to the UK measurements 6 (see discussion in next section). In particular, Nairobi PM has been shown to have a high percentage 7 of mineral dust which typically has a high density, with Gaita et al. (2014) showing the annual 8 average composition of PM 2.5 being composed of 35% mineral dust which originates from unpaved 9 roads and wind-blown dust during the dry seasons. The gravimetric analysis did not allow for the 10 calibration of the PM 1 mass concentrations because a filter sample was not generated for the 11 fraction of PM in this size range. Hence, the PM 1 size fraction calibration uses the same calibration 12 factor derived for the PM 2.5 size fraction. 13 14 The gravimetric calibration was carried out at the urban background field location, for the three 15 OPC-N2s which were subsequently used in the measurement campaign at the three field sites. 16 Hence, the calibration was most appropriate for the urban background site. Whilst the urban 17 roadside site is in close proximity to the urban background site, the roadside site is more influenced 18 by traffic related PM, hence, the average particle density at the roadside site is likely different to the 19 urban background site. Likewise, the rural background site is likely to be far more influenced by 20 mineral dust than the two urban sites. Hence the gravimetric calibration at the urban background 21 sites only provides an estimate calibration for the urban roadside and rural background sites. 22

23
Only one gravimetric calibration was carried out during the study period due to the lack of resource 24 for further calibrations. If the PM composition varied significantly over the study period, then the 25 true calibration factor will also change. Hence, the calibration factor used should be treated as an 26 estimate for the whole study period because changes in PM composition lead to changes in particle 27 refractive index, and therefore, the scattering pattern which is measured by the OPC to estimate 28 particle size. Changes in particle density, due to compositional changes, also affects the particle mass 29 calculated from the particle size. It is noted, for future studies it would be beneficial to have multiple properties to the average rural background. However, PM derived from urban emissions are likely to 10 be less hygroscopic than rural PM; therefore, the rural estimates provide a useful upper estimate of 11 particle hygroscopicity in urban centres. All locations used in the study period typically have RH less 12 than the 85% threshold. However, it is noted that the RH dependent measurements shown in Crilley 13 et al. (2018) were performed in the UK whereas these measurements were performed in Kenya. 14 There may be significant differences between aerosol compositions, and hence hygroscopicities, in 15 these two countries albeit both urban areas (Birmingham and Nairobi) will have significant vehicular 16 influence. Measurements of RH at the Kenyan urban background site show that RH was only equal 17 to or greater than 85% less than 1% of the time. Furthermore, there is no significant dependence of 18 either the observed PM 2.5 or PM 10 mass concentration upon RH (see supplementary figures 1a and  19 1b), this is consistent with low hygroscopicity aerosols. The measurement period of work reported in 20 this paper was in the Kenyan dry season with very few rain events, it is noted that if low cost sensors 21 are to be used in the wet season in Kenya then the RH will likely be greater than 85% during 22 significant periods and the hygroscopicity effect will likely need to be accounted for to obtain good 23 measurements. 24   Continuous monitoring at all three sites was achieved for a fortnight in the period 18/02/2017 to 7 04/03/2017. This period will henceforth be referred to as the intensive period, whereas, the total 8 measurement campaign will be referred to as the campaign period. The number of monitoring days 9

Results
for the urban roadside, urban background and rural background monitoring sites during the 10 campaign period were 40, 29 and 25 days, respectively. 11 12 one third of the time. The urban background site is at an elevated position, which largely removes 1 the direct influence of local sources of PM pollution. As such, it can be assumed that the PM mass 2 concentrations observed at this location represent a lower limit for the ground level PM 3 concentrations throughout Nairobi, since most PM emissions will be due to ground level sources 4 such as vehicle emissions, fires, local industry and others. The rural background site has no daily 5 exceedances in the PM 10 size fraction but exceeds the PM 2.5 guidelines 12% of the time. was observed in all three sites; therefore, it likely represents a long-range pollution event. 10 Correspondingly, the average PM mass concentrations and percentage of WHO exceedances are 11 higher during the intensive period compared to the whole measurement campaign, see Table S1. 12  analysis of the air mass back trajectories indicates that the regional wind direction was almost 5 exclusively from the northeast. Hence the Nanyuki rural background site is a good representative of 6 the rural background that impacts upon Nairobi. 7 The roadside increment was calculated by subtracting the hourly average values of the urban 8 background site from the urban roadside site. It is noted that the chosen roadside measurement 9 site is particularly busy with vehicles, compared to many other non-highway streets in Nairobi. In 10 particular, the site is a popular Matatu (14 seat passenger vans) terminal with multiple vehicles idling 11 at any point during the day. Therefore, the roadside increment obtained using this location likely 12 represents a value close to the upper boundary for Nairobi roads.  The urban and roadside increments are significant for all the investigated PM size fractions. A 5 statistical summary of the roadside and urban increments for the PM 1 , PM 2.5 and PM 10 size fractions 6 are given in Table 3. 7 During the intensive period, the mean average roadside increment is 57.2, 47.5 and 48.1 % of the 11 mean roadside mass concentration, in the PM 1 , PM 2.5 and PM 10 size fractions, respectively. 12 The spatial variation in PM emissions, in the different size fractions, can be assessed at the urban 13 background and urban roadside sites using bivariate polar plots, which provide information on the 14 variation of PM mass concentration with wind direction and speed, see The urban roadside site also shows distinct variation in pollutant concentrations with wind speed 6 and direction. In the PM 10 size fraction the greatest concentrations are seen to the northwest and 7 smallest to the southwest with a steady reduction between these two extremes. The PM 2. 5   and rural background sites which suggests that this site is effected significantly by both the regional 12 background and the urban road PM sources. These insights into the coarse:fine PM ratio is 13 consistent with the roadside and urban increments, shown in Figure 5 and discussed previously. the two studies. The significant increase in measured PM 2.5 could be due to several reasons. Firstly, 10 there could be seasonal differences between August/September and the February/March sampling 11 periods of the two studies; however, the study of Gaita et al. (2014) suggests the urban background 12 concentrations of PM 2.5 mass concentration is similar between these two time periods. The regional 13 background PM loading may have increased during this time period, potentially due to increasing 14 regional aridity caused by climate change leading to more dust generation (Greve et al., 2017). There 15 is almost ten years difference in the times of this study compared to Gaita et al. 2016 PM 2.5 stemmed from the local situation and distinct sources of PM within the two slums. This study 30 used a TSI optical particle counter, which was placed 1.5 m above ground level. Therefore, these 31 measurements were likely highly influenced by re-suspended dust. 98.1 µg/m 3 for a rural and urban roadside site, respectively, compared to ca. 25 µg/m 3 and ca. 150 5 µg/m 3 for this study. Again, the increase between sampling years may be a reflection of the 6 increased population, vehicular traffic and rapid urbanisation. 7

Discussion 8
In this study, we have shown that Nairobi currently has very high levels of PM mass concentration in 9 the PM 1 , PM 2.5 and PM 10 mass fractions. These measurements were conducted using low cost 10 calibrated OPC-N2 sensors. The measured PM 2.5 and PM 10 concentrations at the urban roadside and 11 urban background sites both regularly exceeded the WHO daily limits and very likely exceed the 12 annual limits. In particular, the roadside site often showed concentrations of double the WHO 13 guidelines. These concentrations will very likely be causing significant harm to the population of the 14

Nairobi. 15
The negative health effect of PM is linked to the level of exposure experienced by the patient. This 16 paper and others (e.g. Gaita et al. 2014) have shown that in Nairobi, vehicle emissions are the most 17 significant source of PM. Hence, in Nairobi and other similar cities, the exposure to outdoor PM is to 18 a large extent a function of ones proximity to roads. Furthermore, since traffic varies diurnally, 19 seasonally and by day of the week, personal exposure is both spatially and temporally dependent. 20 This spatial and temporal heterogeneity leads to health inequalities in cities. The urban poor who 21 are often most vulnerable to environmental risks due to lack of adequate health provision, typically 22 live in close proximity to roadways, heightening their exposure to vehicular emissions. Stemming 23 from poorly planned rapid urbanisation and inadequate service provision within these cities, those 24 that are unable to afford public transport or personal vehicles frequently walk along these pollution 25 heavy roads, only increasing their exposure periods. 26 This study only looked at outdoor air quality, it is important to stress that most air pollution deaths 27 in Kenya and SSA in general are due to poor indoor air quality. As a total number of deaths, deaths 28 related to indoor air quality in Kenya rose to 18% from 1990-2013 (Roy, 2016). In LMIC countries, 29 indoor exposure to pollutants is typically from the household combustion of solid fuels on open fires 30 or traditional stoves. These exposures increase the risk of acute lower respiratory infections and 31 associated mortality among young children; indoor air pollution from solid fuel use is also a major 32 This study has shown that the low cost OPC-N2 sensors can be used to generate diurnal PM datasets 4 with good precision and repeatability. As noted in the methodology, it would have been preferable 5 for more cross calibration periods with the tried and tested gravimetric PM measurement but 6 resources did not allow this. In addition to more calibration points, the study could have been 7 enhanced by the inclusion of collocated calibration points for the roadside and rural background 8 sites in addition to the urban background site, since the average particle shape, size and density will 9 likely be different between the three sites because of differing PM sources and emission factors. 10 However, it is noted, that whilst it is desirable from a purely scientific point of view to have more 11 inter-comparison with reference grade equipment; every inter-comparison adds significant 12 additional cost to the project both in terms of consumables for the gravimetric analysis (including 13 the cost of analytical grade filters and accompanying laboratory supplies), and the cost in 14 manpower. Many other cities in SSA and other LMIC countries do not have the resource that Nairobi 15 does in having a gravimetric sampler. These additional costs required for highly accurate scientific 16 results would likely make the low cost sensors not so very low cost after all, and hence bring into 17 question their unique selling point (USP). 18 Whilst this paper focused on PM pollution, it is noted, that there are serious risks to health not only 19 from exposure to PM, but also from exposure to ozone (O 3 ), nitrogen dioxide (NO 2 ) and sulfur 20 dioxide (SO 2 ). As with PM, concentrations are often highest largely in the urban areas of low-and 21 middle-income countries. Ozone is a major factor in asthma morbidity and mortality, while nitrogen 22 dioxide and sulfur dioxide also can play a role in asthma, bronchial symptoms, lung inflammation 23 and reduced lung function. Good quality measurements of these gas phase pollutants lag behind 24 measurements of PM in Nairobi, other SSA cities and LMIC cities in general. This is due to the high 25 importance of PM as an environmental risk factor but also because of the lack of good quality gas 26 analysers which are affordable and transportable. 27 28

Conclusions 29
Air quality in many LMIC urban centres is often poor and in many cities is getting worse due to the 30 combined pressures of increasing population, increasing urbanization, increasing vehicular traffic 31 and poor vehicle regulation. To be able to manage air pollution, good quality and long term data sets 32 are required. Unfortunately, in many LMICs the cost of certified high quality air quality 1 measurements is beyond the financial means of environmental authorities. Low cost sensors offer 2 the possibility of air quality products at significantly lower cost compared to traditional methods. 3 This paper used calibrated OPC-N2 devices to measure PM concentrations in Nairobi, Kenya in the 4 size fractions PM 1 , PM 2.5 and PM 10 . The data required calibration using an established gravimetric 5 approach to PM measurement. The need for calibration by trained scientists significantly increases 6 the costs associated with low cost monitoring and this cost needs to be factored in when assessing 7 options for air quality monitoring. 8 PM was measured in three locations: an urban roadside, urban background and rural background 9 site for a period of approximately two months. The data reveals that roadside and urban 10 background locations in Nairobi often exceed the WHO guidelines for daily averaged PM mass 11 concentration in both the PM 2.5 and PM 10 size fractions. Comparison of the data with previous 12 measurements conducted in Nairobi is difficult but where comparison is possible, it appears that air 13 quality has become worse in the last ten years which is likely due to increases in population, 14 urbanisation and motorization. Comparison of the data from the three sites, following a 'Lenschow ' 15 type approach, allowed for the calculation of representative roadside and urban increments for 16 Nairobi (Lenschow et al., 2001). This increment data can be used in future air quality modelling to 17 assess the likely health impact of PM pollution on Nairobi's population. The combination of the 18 diurnal PM data with local meteorology allows for simple source apportionment of the PM. The 19 diurnal PM concentrations tracks the Nairobi rush hours, furthermore, PM peaks when the wind 20 comes from the direction of significant numbers of vehicles such as major roads and a Matatu stop. 21 These facts taken together, point towards vehicle emissions being the major sources of air pollution 22 in Nairobi, as has been previously observed in studies such as Gaita et al. 2014. The coarse PM 23 fraction increases at roadside compared to urban background site suggesting that non-exhaust 24 vehicle emissions make up a significant amount of the vehicle emissions. 25 In summary, the low cost sensors used in this study provided much useful data for assessing air 26 quality in Nairobi at an equipment cost significantly lower than that of traditional instruments. Low 27 cost sensors have great potential in other country settings and could be used for long term sampling 28 if the appropriate calibrations are performed.