Airborne particulate matter monitoring in Kenya using calibrated low cost sensors

East African countries face an increasing threat from poor air quality stemming from rapid urbanization, population growth, and a steep rise in fuel use and motorization rates. With few air quality monitoring systems available, this study provides much needed high temporal resolution data to investigate the concentrations of particulate matter (PM) air pollution in Kenya. Calibrated low-cost optical particle counters (OPCs) were deployed in Kenya in three locations: two in the capital Nairobi and one in a rural location in the outskirts of Nanyuki, which is upwind of Nairobi. The two Nairobi sites consist of an urban background site and a roadside site. The instruments were composed of an AlphaSense OPC-N2 ran with a Raspberry Pi low-cost microcomputer, packaged in a weather-proof box. Measurements were conducted over a 2-month period (February–March 2017) with an intensive study period when all measurements were active at all sites lasting 2 weeks. When collocated, the three OPCN2 instruments demonstrated good inter-instrument precision with a coefficient of variance of 8.8± 2.0 % in the fine particle fraction (PM2.5). The low-cost sensors had an absolute PM mass concentration calibration using a collocated gravimetric measurement at the urban background site in Nairobi. The mean daily PM1 mass concentration measured at the urban roadside, urban background and rural background sites were 23.9, 16.1 and 8.8 μg m−3, respectively. The mean daily PM2.5 mass concentration measured at the urban roadside, urban background and rural background sites were 36.6, 24.8 and 13.0 μg m−3, respectively. The mean daily PM10 mass concentration measured at the urban roadside, urban background and rural background sites were 93.7, 53.0 and 19.5 μg m−3, respectively. The urban measurements in Nairobi showed that PM concentrations regularly exceed WHO guidelines in both the PM10 and PM2.5 size ranges. Following a “Lenschow”-type approach we can estimate the urban and roadside increments that are applicable to Nairobi (Lenschow et al., 2001). The median urban increment is 33.1 μg m−3 and the median roadside increment is 43.3 μg m−3 for PM2.5. For PM1, the median urban increment is 4.7 μg m−3 and the median roadside increment is 12.6 μg m−3. These increments highlight the importance of both the urban and roadside increments to urban air pollution in Nairobi. A clear diurnal behaviour in PM mass concentration was observed at both urban sites, which peaks during the morning and evening Nairobi rush hours; this was consistent with the high roadside increment indicating that vehicular traffic is a dominant source of PM in the city, accounting for approximately 48.1 %, 47.5 % and 57.2 % of the total PM loading in the PM10, PM2.5 and PM1 size ranges, respectively. Collocated meteorological measurements at the urban sites were collected, allowing for an understanding of the location of major sources of particulate matter at the two sites. The potential problems of using low-cost sensors for PM measurement without gravimetric calibration available at all sites are discussed. This study shows that calibrated low-cost sensors can be successfully used to measure air pollution in cities like Nairobi. It demonstrates that low-cost sensors could be used to create an affordable and reliable network to monitor air quality in cities. Published by Copernicus Publications on behalf of the European Geosciences Union. 15404 F. D. Pope et al.: Airborne particulate matter monitoring in Kenya using calibrated low-cost sensors

53.0 and 19.5 µg m −3 , respectively.The urban measurements in Nairobi showed that PM concentrations regularly exceed WHO guidelines in both the PM 10 and PM 2.5 size ranges.Following a "Lenschow"-type approach we can estimate the urban and roadside increments that are applicable to Nairobi (Lenschow et al., 2001).The median urban increment is 33.1 µg m −3 and the median roadside increment is 43.3 µg m −3 for PM 2.5 .For PM 1 , the median urban increment is 4.7 µg m −3 and the median roadside increment is 12.6 µg m −3 .These increments highlight the importance of both the urban and roadside increments to urban air pollution in Nairobi.
A clear diurnal behaviour in PM mass concentration was observed at both urban sites, which peaks during the morning and evening Nairobi rush hours; this was consistent with the high roadside increment indicating that vehicular traffic is a dominant source of PM in the city, accounting for approximately 48.1 %, 47.5 % and 57.2 % of the total PM loading in the PM 10 , PM 2.5 and PM 1 size ranges, respectively.Collocated meteorological measurements at the urban sites were collected, allowing for an understanding of the location of major sources of particulate matter at the two sites.The potential problems of using low-cost sensors for PM measurement without gravimetric calibration available at all sites are discussed.
This study shows that calibrated low-cost sensors can be successfully used to measure air pollution in cities like Nairobi.It demonstrates that low-cost sensors could be used to create an affordable and reliable network to monitor air quality in cities.

Introduction
Recently, the Lancet Commission on pollution and health estimated that in 2015, air pollution led to the premature deaths of over 9 million people globally, and contributed to over one in four deaths in severely affected countries (Landrigan et al., 2017).Typically, urban air pollution is higher in lowand middle-income countries (LMICs) compared to further developed countries.Hence, the associated risk of air pollution to health is typically higher in LMICs, with over 92 % of global pollution-related deaths occurring in these countries.Within LMICs, health inequalities in urban areas contribute to an increased exposure to air pollution that expose those that live, work, socialize and commute to highly urbanized areas, which typically have a substantially higher concentration of air pollutants.Despite the extensive links between air pollutants and human health, environmental degradation, and the economy, air pollution is as of yet still under-researched in many LMICs.Due to a lack of long-term air-quality monitoring in many LMICs, the concentrations and sources of air pollution are poorly understood.
Airborne particulate matter (PM) is a major environmental risk factor with well-documented short-and longterm effects on human mortality and morbidity (Thurston et al., 2016).Long-term side effects to air pollution exposure include asthma, chronic obstructive pulmonary disease (COPD), pulmonary fibrosis, cancer, type 2 diabetes, neurodegenerative diseases, obesity and other conditions (Ferranti et al., 2017;Landrigan et al., 2017).People who are already susceptible to underlying respiratory disease, pneumonia, influenza or asthma can see a worsening of their symptoms and illnesses through short-term exposure to air pollution (Wan Mahiyuddin et al., 2013).The size of PM is correlated with their health impacts, with smaller particles typically having more significant health implications (Meng et al., 2013).PM 1 , PM 2.5 and PM 10 are particulate matter with aerodynamic diameters less than 1, 2.5 and 10 µm, respectively.The World Health Organization (WHO) recommends that PM 2.5 and PM 10 daily mass concentrations should not exceed 25 and 50 µg m −3 , respectively; and annual mass concentrations do not exceed 10 and 20 µg m −3 , respectively (WHO, 2006).At present, the WHO or other regulatory bodies do not provide recommendations of the mass concentrations of PM 1 .PM 1 can remain suspended in air for much longer than coarser PM, as well as penetrating deeper into the lungs leading to local pulmonary and systematic inflammation (Pateraki et al., 2014).Due to the smaller size, PM 1 has a higher surface to mass ratio, containing a harmful amount of potentially toxic anthropogenic constituents which could lead to health impacts such as respiratory disease, heart disease and lung cancer (Trippetta et al., 2016).Many studies still focus on PM 10 and PM 2.5 even though smaller particulates pose greater health impacts (Tsiouri et al., 2015).Beyond PM 1 , ultra-fine particles (< 100 nm) are of such a small size that they can be translocated to the central nervous sys-tem via the blood-brain barrier or the olfactory bulb (Chen et al., 2016).There are no air quality regulations for PM 1 or ultra-fine particles due to the paucity of data either within environmental science or public health.
Worldwide, road traffic is a dominant source of urban PM accounting for 5 %-80 % of PM mass, with the precise amount being dependent upon several factors including time, location and vehicle fleet, as reviewed by Pant and Harrison (2013).Vehicle-derived PM is directly associated with negative health outcomes (Fan et al., 2006;HEI, 2010).Emissions are due both to both exhaust-pipe emissions and non-exhaust-pipe emissions.Exhaust emissions result from the combustion of fuel, predominantly petrol and diesel, such as oil and other lubricants.Non-exhaust emissions come either from the resuspension of road dust through wind-or vehicle-induced wind shear, or from the wear and tear of vehicle parts including the brakes, tyres and clutch.Resuspension of dust is particularly important on non-paved roads of which there are an abundance in Nairobi.Typically, nonexhaust emissions are in the coarse PM size fraction (PM in the size range 2.5-10 µm aerodynamic diameter), whereas exhaust emissions are in the fine PM size fraction (PM 2.5 ) (Thorpe et al., 2007;Kam et al., 2012).However, it is noted that the papers which reference vehicle PM size distributions according to the emission of non-exhaust sources have typically been conducted in either the US or European studies and not in Africa, where non-paved road sources represent a much higher fraction of road surface types.The precise size of vehicular-derived PM is dependent on several factors: vehicle fleet characteristics (e.g.weight and size), road type, level of maintenance and meteorological conditions (Beddows et al., 2009;Hays et al., 2011).
In many cities of LMICs, urbanization, population, fuel use and motorization rates are all increasing rapidly and increases in air pollution are associated with these trends (Mitlin and Satterhwaite, 2013;Ochieng et al., 2017).In particular, vehicular traffic is fast on the rise, with associated congestion on the road networks, which can contribute as much as 90 % to air pollution in urban environments (Kinney et al., 2011).Nairobi is the capital city of Kenya and is showing these trends.In particular, the city population has increased dramatically, since 1999 to 2015 it has risen by 83 %, and is projected to increase to 7.14 million by 2030 (Rajé et al., 2017).Similarly, motorization rates are increasing; between 2008 and 2012, the number of motor-and auto-cycles in Kenya grew by 368 % with the number of overall registered vehicles increasing by 77 % (Rajé et al., 2017).Considering this extensive increase in the vehicle fleet, limited roadway infrastructure and high congestion within the city, pollution hotspots are created leading to personal exposure levels much higher than that encountered throughout the rest of the city (van Vliet and Kinney, 2007).
To be able to reduce air pollution, it is helpful to be able to measure it, so reduction efforts can be assessed.Many LMICs have insufficient monitoring networks through which to measure air quality.In particular, long-term highresolution data are required for such cities which are vulnerable to air pollution.Nairobi is in the vanguard of air pollution measurements for sub-Saharan Africa (SSA) but lacks continuous long-term calibrated measurements of PM and other air pollutants (Petkova et al., 2013).A discussion of the relevant measurements in Nairobi is given in the next section.One of the constraints to making measurements is the high cost of research-grade air quality monitoring equipment with appropriate calibration and certification.Low-cost sensors offer the potential for dramatically reducing equipment costs by orders of magnitude, making the monitoring of air quality more accessible and attainable in LMICs (Lewis et al., 2016;Rai et al., 2017).
In this paper, the use of low-cost sensors for measurement of PM 1 , PM 2.5 and PM 10 in Nairobi is detailed.We have previously assessed the same low-cost sensors in the UK (Crilley et al., 2018).The sensors are calibrated using a standardized gravimetric approach.PM is measured in three locations: an urban roadside site, an urban background site and a rural background site.Comparison of simultaneous measurements at the three sites allows for the estimation of an urban increment and roadside increment in PM following a Lenschow-type approach, which allows for simple modelling of city air pollution based on the urban and roadside increments in air pollution (Lenschow et al., 2001).The variation in measured PM with measured meteorological data is also discussed.Finally, we discuss the implications of using lowcost sensors in Nairobi and LMICs in general.

Previous PM measurements in Nairobi
In general, long-term air-quality monitoring in sub-Saharan Africa (SSA) is rare.Correspondingly, there are only limited PM data sets for East African urban areas; where data does exist, estimated concentrations for PM 2.5 concentrations are approximately 100 µg m −3 compared to < 20 µg m −3 in most European and North American cities (Brauer et al., 2012).This indicates that urban PM air pollution in East Africa could be a significant health concern.
In Nairobi, there have been numerous short-term measurements of PM over the last decade (Brauer et al., 2012;Kinney et al., 2011;Ngo et al., 2015;Egondi et al., 2016;Gaita et al., 2016) with only one long-term continuous measurement (Gaita et al., 2014).To date, most measurements have used gravimetric measurement methodologies to record PM mass concentration in the PM 2.5 and PM 10 size fractions.Most measurements indicate that PM concentrations in Nairobi regularly exceed the WHO guidelines.At present, there is only one publication in the scientific literature describing the use of low-cost sensors in the measurement of PM (deSouza et al., 2017) that monitored air quality in Nairobi at six sites from May 2016 to January 2017.Using AlphaSense OPC-N2 instruments, the authors measured PM 1 , PM 2.5 and PM 10 , as well as NO 2 , NO and SO 2 .
The study collected PM concentrations at six schools within Nairobi.It reported a PM 2.5 concentration range between 11 and 21 µg m −3 , and a range of 26 to 59 µg m −3 for PM 10 .The PM concentrations measured during the deSouza et al., 2017, study are noticeably lower than this study for both size fractions.It is worthy to note that the deSouza study collected measurements from May 2016 to January 2017, whereas this campaign took place from February to April 2017; the local meteorology may have influenced the discrepancies seen in both recorded PM concentrations.Additionally, the study did not calibrate the monitors, which leads to questions about absolute concentrations and interference from other environmental dependencies (Lewis and Edwards, 2016).The collected data from the study appeared noisy, with the authors stating they could not separate the signal from the noise without having access to an air quality measuring reference instrument (they recorded peaks at over 1000 µg m 3 ).Despite the limitations, it provides a useful comparison to this calibrated study.
The paucity of long-term calibrated measurements has hindered the understanding of long-term trends and the influence of seasonal variations in meteorology and other factors.Most published data provide daily averages of PM mass; the lack of higher temporal resolution data precludes the generation of diurnal data, which can be useful for identifying individual sources of PM, in particular, vehicular PM that typically tracks traffic and hence peaks during rush hours.
The longest record of PM concentration in Nairobi is detailed in Gaita et al. (2014).In that work, the authors performed daily measurements of PM 2.5 at an urban background and suburban site over a 2-year period from May 2008 to April 2010 using polycarbonate filters in a cyclone sampler.They reported a concentration range of 3 to 53 µg m −3 at the urban background site, with an overall mean of 21 µg m −3 , which exceeds the annual WHO limit of 10 µg m −3 by a factor of 2. The average concentrations of PM 2.5 at both sites were found to be 21 ± 9.5 and 13 ± 7.3 µg m −3 , respectively.Chemical composition measurements of the filter samples allowed source apportionment, via positive matrix factorization, to be carried out.The analysis suggested that five major source factors contribute to Nairobi PM 2.5 : traffic, mineral dust, industry, combustion and a mixed factor.The dominant source factors were mineral dust and traffic, which accounted for 74 % of the particle mass.
As an update to this study, Gaita et al. (2016) conducted a study on the characterization and size fractionation of PM and deposition fraction in the human respiratory system in Nairobi using measurements taken in August and September 2007, obtained at the University of Nairobi site.Based on the findings, the concentration levels of airborne PM sampled at the urban background site during the period was found to range between 1 and 78 µg m −3 .The average PM 2.5 con-centration at the site over the entire sampling period was 9.8 ± 8.5 µg m −3 .
A densely populated urban area with associated heavy local traffic within Nairobi largely contributes to the city's air pollution build-up.Kinney et al. (2011) investigated the impact of vehicular emissions in Nairobi on the concentration of PM 2.5 , observing a substantial range between 58 and 98 µg m −3 across an 11 h personal exposure along busy roadways and roundabouts.The range was estimated to be between 45 and 85 µg m −3 for a 24 h sampling due to pollutant dispersion at night.In addition, the study reported a decrease in horizontal dispersion measurements of PM 2.5 from 128.7 to 18.7 µg m −3 over 100 m downwind of a major intersection in Nairobi.A vertical dispersion from street level to a third-floor rooftop in the Central Business District (CBD) showed a decrease in PM 2.5 concentration from 119.5 to 42.8 µg m −3 .This study clearly highlights that the PM concentration in Nairobi varies considerably over both time and space, which has significant implications for human exposure (see Discussion section).
Another study by Ngo et al. (2015) affirmed the contribution of anthropogenic activities on the quality of air in Nairobi.In their study, Teflon filters in PM 2.5 samplers (BGI model 400) were used between 2 and 18 August 2011 and high concentrations of PM 2.5 exposure levels among different groups in Nairobi were reported.According to the study, bus drivers in Nairobi city were exposed to about 103 µg m −3 , while those in informal settlements, such as Mathare, reporting exposure levels of about 62.7 µg m −3 , an indication that urgent measures are needed to be taken to mitigate the impact of air pollution in the city.
The severity of air pollution in urban centres in SSA is typically even higher in the informal settlements (slums), where acute respiratory tract infections and bronchitis are among the most frequent medical diagnoses (Gulis et al., 2004).Egondi et al. (2016), in their study on air pollution in two informal settlements in Nairobi: Korogocho and Viwandani, reported higher levels of PM 2.5 in the two slums.Optical counters (TSI DustTrak II model 3530) were used in the study and observed average levels of PM 2.5 in Korogocho slum, lying west of Dandora, Nairobi's biggest dumping ground, were the highest at 166 µg m −3 and Viwandani, situated north of Nairobi, recorded 67 µg m −3 .

Site locations
This study used three field sites in Kenya (see Fig. 1).Two sites were in Nairobi, which is the capital of Kenya, covering an area of ca.696 km 2 and home to approximately 3.5 million residents according to a World Population review conducted in 2016 (Kenya Population, 2016), making it the second most populated city in East Africa after Dar es Salaam, Tanzania.In addition to Nairobi's longstanding popularity as a travel destination, due to its safari and other holiday resorts, the city also acts as East Africa's diplomatic, financial and communication capital (Rajé et al., 2017).
Its geographical location is at approximately 1.29 • S and 36.82• E. The highest elevation point in the city is at an altitude of 1663 m above the ground.As discussed in the introduction, Nairobi is undergoing rapid increases in population and motorization both of which will likely lead to greater PM pollution in the absence of any efforts of mitigation against the pollution.Other significant infrastructure projects such as major road building are currently being undertaken, which will also likely lead to increased PM loadings.Within Nairobi, the two field sites represent an urban background location and an urban roadside location.The other site, a rural background site, is located on the outskirts of Nanyuki, a town that is located at an approximate aerial distance of 147 km to the north-northeast of Nairobi and 240 km by road.The sensor boxes were placed in locations free from obstacles, at the three measurement sites, allowing for 360 • of air flow.A description of the meteorological conditions is provided in the Supplement.

Site 1: American Wing, University of Nairobi, Nairobi (urban background site)
The first site for data collection in Nairobi was at the American Wing building located in the University of Nairobi, standing at an elevation of 17 m a.g.l.Air flow at the site was free from any obstruction as the OPCs were located at an elevated point above the ground.The nearest road is Harry Thuku Road, which has very few on-road vehicles (no heavy trucks) and it leads to Fairmont The Norfolk and Boulevard hotels, and Kijabe Street.
3.1.2Site 2: Tom Mboya Street, fire station, Nairobi (urban roadside) The second collection site in Nairobi was at the fire station, which is located within the CBD in the city.Unlike the American Wing site, the area around the fire station is characterized by high traffic flow which includes common public transport vans, locally known by the name "matatus".It is on an urban street canyon, on a street where smoking diesel vans are frequent and is exposed to urban heat island effects.It is also in the neighbourhood of a vertical dispersion measurement site of PM 2.5 used by Kinney et al. (2011).The monitor was mounted at a height of approximately 5 m.

Site 3: Nanyuki town (rural background)
The third site chosen was on the outskirts of Nanyuki town, an administrative town in Laikipia County, which is located to the northwest of Mt Kenya.The town is positioned at the Equator at approximately 1.28  , 2015).The Nanyuki region has a hinterland of significant agricultural cultivation, forest and considerable grazing activities (Gatari et al., 2005).The OPC was hung about 4 m a.g.l.thus exposing it to free regional air mass in an area of minimal local influence.

PM measurement equipment
Small low-cost optical particle sensors ( The assumed density for all particle sizes is 1.65 g cm −3 and no special weighting is placed on any particular bin size.However, the manual for the OPC states "an additional weighting is applied on units with Firmware 18 or higher to account for under counting at low particle sizes and the effect of carbon particles in urban air so that the output matches collocated reference detectors." The lower cut-off for particle size observed by the OPC is 380 nm and hence a large proportion of all particles are not observed by the OPC due to the particle number being dominated by the smallest particle sizes (Seinfield and Sypyros, 2016).Ultrafine particles (particles of aerodynamic diameter < 100 nm) were therefore not measured.However, the interest of the study was particulate mass which is dominated by particle sizes that were measured.
The sensors had their data logged using Raspberry Pi 3 minicomputers.The Python script used to run the OPC-N2 on the Raspberry Pi 3 is discussed and provided in Crilley et al. (2018) and makes use of the py-opc python library for operating the OPC-N2 written by Hagan (2017).Together, the OPC-N2, minicomputer and accompanying wires and tubing were placed in bespoke weather-resistant housing (dimensions ca.30 cm× 20 cm× 10 cm).Power for both the OPC-N2 and minicomputer were provided by mains power.
The OPC-N2 sensors are factory calibrated to measure PM mass concentrations representative of the UK.However, in our previous study (Crilley et al., 2018), we demonstrated that in situ calibration of the sensors is required for the correct measurement of PM mass concentrations at urban background sites in Birmingham, UK.The calibration in the Cril-ley et al. ( 2018) study involved both scaling and a relative humidity (RH) dependent term for when the RH is greater than approximately 85 %.
The mass concentrations from the OPC-N2 devices, in the PM 1 , PM 2.5 and PM 10 size bins were recorded in time intervals of 10 s.For the subsequent analysis, the mass concentration data were aggregated into 1 h time bins using the mean average.In time periods which contained missing data, the mean average of the available data was aggregated.All data manipulations were performed using R (version 3.4.1),and the openair project package for R was extensively used for data visualization (Carslaw and Ropkins, 2012).

Meteorological station
The local meteorology for Nairobi was measured at the same location as the urban background site using a Vaisala instrument (WXT510) with the following variables measured: wind speed, wind direction, temperature, relative humidity, barometric pressure and rainfall with an instrument temporal resolution of 5 minutes.The measured meteorology parameters were in good agreement with other local measurements such as those observed at Jomo Kenyatta International Airport (JKIA), which is approximately at an aerial distance of 10 km.The proximity of the meteorological station at the urban background site to the urban roadside makes the meteorological data appropriate for both sites.The data were collected at the urban background site from the 2 February to the 6 April 2017, covering the duration of the PM measurements.

OPC-N2 gravimetric mass calibration
The OPC-N2 mass concentrations were calibrated using gravimetric measurements of PM 2.5 and PM 10 .The gravimetric calibration measurement was carried out on the 9 February 2017 for 24 h.A collocation measurement of the OPC and an Anderson dichotomous impactor (Sierra Instruments Inc., USA) was set up, on the only possible date, at the background site.The impactor collected PM 2.5 and PM 10−2.5 particles on Teflon filters (diameter = 37 mm, pore size = 2 µm) at a total flow rate of 1 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 was rain free and had similar temperature and RH profiles compared to the rest of the OPC sampling campaign.The filters were weighed using a mass balance before and after PM collection.The observed 24 h average mass concentrations of PM 2.5 and PM 10 from the impactor were 27.6 ± 6.8 and 51.8 ± 10.3 µg m −3 , respectively, while those recorded from the OPC were 16.9 and 30.6 µg m −3 , respectively.The uncertainty in gravimetric concentrations was estimated from the instrument (10 %), sampling (7 %) and weighing (25 %) errors, and that of the OPC data was the standard deviation.The value is as provided by the OPC, it is not possible to provide an error from the 1-day calibration.From Crilley et al. (2018)  The gravimetric calibration was carried out at the urban background field location for the three OPC-N2s which were subsequently used in the measurement campaign at the three field sites.Hence, the calibration was most appropriate for the urban background site.Whilst the urban roadside site is in close proximity to the urban background site, the roadside site is more influenced by traffic-related PM; hence, the average particle density at the roadside site is likely different to the urban background site.Likewise, the rural background site is likely to be far more influenced by mineral dust than the two urban sites.Hence the gravimetric calibration at the urban background sites only provides an estimate calibration for the urban roadside and rural background sites.
Only one gravimetric calibration was carried out during the study period due to the lack of resources for further calibrations.If the PM composition varied significantly over the study period, then the true calibration factor will also change.Hence, the calibration factor used should be treated as an estimate for the whole study period because changes in PM composition lead to changes in particle refractive index, and therefore the scattering pattern which is measured by the OPC to estimate particle size.Changes in particle density, due to compositional changes, also affects the particle mass calculated from the particle size.It is noted that for future studies it would be beneficial to have multiple gravimetric calibration points to check for continuing accuracy of the OPC-N2 sensors throughout the campaign.

Measured particle mass dependence on relative humidity
As detailed in Crilley et al. (2018), under UK conditions, the OPC-N2 device is sensitive to variations in RH when the RH exceeds ca.85 %.Crilley et al. (2018) suggest the RH dependence is due to the hygroscopic properties of particles that result in significant water mass being taken up by PM at high RH.This hygroscopic dependence can be modelled using a calibration that uses the κ-Kohler parameterization of aerosol hygroscopicity (Petters and Kreidenweis, 2008).The average κ parameter values for Africa (κ = 0.15 ± 0.12) are lower than for Europe (κ = 0.36 ± 0.16), as based on the Pringle et al. ( 2010) model, which is in good agreement with observational data.It is noted that composition of urban PM will have different hygroscopic properties to the average rural background.However, PM derived from urban emissions are often less hygroscopic than rural PM; therefore, the rural estimates might provide a useful upper estimate of particle hygroscopicity in urban centres.All locations used in the study period typically have RH less than the 85 % threshold.However, it is noted that the RH dependent measurements shown in Crilley et al. (2018) were performed in the UK, whereas these measurements were performed in Kenya.There may be significant differences between aerosol compositions, and hence hygroscopicities, in these two countries albeit both urban areas (Birmingham and Nairobi) will have significant vehicular influence.Measurements of RH at the Kenyan urban background site show that RH was only equal to or greater than 85 % less than 1 % of the time.Furthermore, there is no significant dependence of either the observed PM 2.5 or PM 10 mass concentration upon RH (see Supplement Fig. S1a and   b), this is consistent with low hygroscopicity aerosols.The measurement period of work reported in this paper was in the Kenyan dry season with very few rain events, it is noted that if low-cost sensors are to be used in the wet season in Kenya then the RH will probably be greater than 85 % during significant periods and the hygroscopicity effect will likely need to be accounted for to obtain good measurements.

Particulate matter measurement
PM data were collected at the three sampling sites over the time period inclusive of 2 February and 24 March 2017.Figure 2 provides the time series data for the PM 10 , PM 2.5 and PM 1 data over the whole measurement campaign.Gaps in data at specific sites are either due to the co-location of two or all three instruments at one site for cross calibration purposes, due to power failure requiring instrument restart or OPC malfunctioning.
The inter-instrument OPC-N2 precision was measured once during the campaign by co-locating the three instruments at the urban background site for 3 days for side by side sampling.The three instrument co-location was carried out during at the start of the campaign (16-18 February 2017).Two OPCs were collocated together at the urban roadside site near the end of the campaign (4-27 March 2017).All instruments gave very similar readings during both co-location periods, the inter-instrument precision gave a coefficient of variance of 8.8 ± 2.0 % in the PM 2.5 fraction, with no degradation in inter-instrument precision observed over the sampling period.This coefficient of variance is better than observed in Crilley et al. (2018) but this is expected because of the lower RH conditions in Nairobi (see previous discussion).
Continuous monitoring at all three sites was achieved for a fortnight in the period 18 February to 4 March 2017.This period will henceforth be referred to as the intensive period, whereas the total measurement campaign will be referred to as the campaign period.The number of monitoring days for the urban roadside, urban background and rural background monitoring sites during the campaign period were 40, 29 and 25 days, respectively.
Table 1 provides the average PM 1 , PM 2.5 and PM 10 mass concentrations observed at the three sites during the campaign period.An identical table for the intensive period is included in the Supplement (see Table S1).The percentage of daily exceedances of daily PM 2.5 and PM 10 as per WHO guidelines are also provided; however, to date there is no set guidelines of PM 1 .All measurement sites exceeded the WHO daily guidelines for both PM 2.5 and PM 10 for some of the days sampled.The urban roadside site exceeds the WHO guidelines on most days (85 % for PM 2.5 and 90 % for PM 10 ).Furthermore, on many days (13 % for PM 2.5 and 40 % for PM 10 ) the urban roadside site exceeds the WHO www.atmos-chem-phys.net/18/15403/2018/guidelines by at least twice as much.The urban background site has fewer exceedances, compared to the urban roadside site, with daily exceedances occurring approximately onethird of the time.The urban background site is at an elevated position, which largely removes the direct influence of local sources of PM pollution.As such, it can be assumed that the PM mass concentrations observed at this location represent a lower limit for the ground level PM concentrations throughout Nairobi, since most PM emissions will be due to ground level sources such as vehicle emissions, fires, local industry and others.The rural background site has no daily exceedances in the PM 10 size fraction but exceeds the PM 2.5 guidelines 12 % of the time.
During the 2-week intensive campaign, there was a period of elevated PM mass concentration observed in PM 1 , PM 2.5 and PM 10 size fractions centred around the 23 February.The elevated PM was observed in all three sites; therefore, this might represent a long-range pollution event.Correspondingly, the average PM mass concentrations and percentage of WHO exceedances are higher during the intensive period compared to the whole measurement campaign (see Table S1).
Whilst there is insufficient temporal data to provide a yearly average value for PM 2.5 and PM 10 mass concentrations for the three sites, the annual average values can be estimated from the data set using the average values provided in Table 1.These values are likely to be upper esti-mates for the yearly values because the measurements were obtained in the period with little precipitation, thereby minimizing the degree of wet deposition of the PM.For instance, Gaita et al. (2014) showed that Nairobi's short rainy season (typically October-December) suppresses PM concentrations at the urban background site by approximately 50 %.Notwithstanding the seasonal rain consideration, the average PM mass concentration observed in this study suggests that the WHO recommendations for annual PM 2.5 and PM 10 are likely exceeded at both the urban background and urban roadside locations.For the urban background site, the measured average PM 2.5 and PM 10 mass concentrations exceed the annual WHO recommendations by factors of 2.5 and 2.7, respectively.Whereas for the urban roadside site they exceed recommendations by 3.7 and 4.7, respectively.These significant exceedances for both the urban roadside and urban background sites suggests that most of Nairobi's population will be subjected to outdoor air pollution far in excess of the WHO recommendations for annual exposure.Figure 3 provides the box and whisker plots for the hourly averaged PM 2.5 and PM 10 data for the three measurement sites.Figure S3 provides box and whisker plots for the daily averaged PM 2.5 and PM 10 data for the three measurement sites and highlights the proportion of the days which exceed the WHO annual and daily recommendations.
Figure 4 provides the mean average diurnal hourly profiles of the PM 1 , PM 2.5 and PM 10 mass concentrations for  the three measurement sites during the whole campaign period.There is clear diurnal variation observed at all the sites, two distinct peaks are observed in the two urban locations during the morning (ca.05:00-10:00 EAT) and the evening (ca.18:00-24:00), which correspond to the Nairobi peak traffic periods.The normalized data shows that the traffic-related structure is very similar in both the urban background and urban roadside sites indicating that the traffic-related PM pollution is the dominant source at both sites.The rural back-ground site also shows diurnal variation with some indication of a traffic-related signal at similar times to the urban sites, especially in the PM 2.5 size fraction.However, overall the rural diurnal cycle appears to largely correspond to solar insolation suggesting the dominant factor affecting the rural mass concentrations is the height of the local boundary layer, which decreases in the night time and increases with greater solar insolation.Through comparison of the urban roadside, urban background and rural background hourly averaged data, it is possible to generate estimates of urban increments and roadside increments relevant for Nairobi using a Lenschow-type approach (Lenschow et al., 2001).For the intensive period, the urban and rural increments are calculated for the PM 1 , PM 2.5 and PM 10 mass concentrations (see Fig. 5).The urban increment is calculated by subtracting the hourly average values of the rural background site from the urban background site.During the intensive period, analysis of the air mass backtrajectories indicates that the regional wind direction was almost exclusively from the northeast.Hence the Nanyuki rural background site is a good representative of the rural background that impacts upon Nairobi.
The roadside increment was calculated by subtracting the hourly average values of the urban background site from the urban roadside site.It is noted that the chosen roadside measurement site is particularly busy with vehicles, compared to many other non-highway streets in Nairobi.In particular, the site is a popular matatu (14 seat passenger vans) terminal with multiple vehicles idling at any point during the day.Therefore, the roadside increment obtained using this location likely 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 statistical summary of the roadside and urban increments for the PM 1 , PM 2.5 and PM 10 size fractions are given in Table 2.
During the intensive period, the mean average roadside increment is 57.2 %, 47.5 % and 48.1 % of the mean roadside mass concentration, in the PM 1 , PM 2.5 and PM 10 size fractions, respectively.
The spatial variation in PM emissions, in the different size fractions, can be assessed at the urban background and ur-  ban roadside sites using bivariate polar plots, which provide information on the variation in PM mass concentration with wind direction and speed (see Fig. 6; Carslaw and Beevers, 2013).The urban background and urban roadside sites are sufficiently closely collocated (< 0.5 km apart) that the wind data acquired at the urban background site is applicable to the urban roadside site.Wind direction data were not available for the rural background site, so analysis of the spatial variation was impossible at this site.Figure 6 clearly shows significant variation in PM mass concentration at both urban sites, which are dependent upon the wind conditions.The urban background site shows broadly similar behaviour in the spatial variation in the PM 1 , PM 2.5 and PM 10 size fractions.The peak in concentrations are observed at low wind speeds and when the wind comes from the west and south.This wind direction dependence is consistent with the close proximity of the major highway A104 "Nairobi-Malaba Road", which passes close to the site in the direction of high PM concentrations.The diurnal profiles and roadside increments discussed earlier combined with the wind dependence highlights the role of roads in Nairobi as the major source of PM in all size fractions studied.Since the site is within Nairobi's CBD, there are other significant roads nearby as well, but the A104 has the greatest fleet density.
The urban roadside site also shows distinct variation in pollutant concentrations with wind speed and direction.In the PM 10 size fraction, the greatest concentrations are seen to the northwest and smallest to the southwest with a steady reduction between these two extremes.The PM 2.5 and PM 1 size fractions show a more complex behaviour with highest concentrations at low wind speeds and the north and west directions.The urban roadside location is surrounded by small roads and lower traffic speeds compared to the highways, for example the A104.The lower traffic speeds likely lead to less non-exhaust-pipe emissions from dust resuspension and hence there are less local PM 10 particles when compared to the urban background site.Whereas the localized PM 2.5 and PM 1 concentrations are likely due to the heavily congested local roads on which matatus and other vehicles are often left idling leading to high exhaust-pipe emissions, which are typically in the smaller PM size fractions (Pant and Harrison, 2013).
Figure 7 provides the distributions of the ratio between the coarse and the fine PM mass fractions at the three field sites during the campaign period.Whilst each site shows distinct variation (with large interquartile ranges) in the reported coarse to fine ratio -which is dependent on the time of year and time of day -the median ratios at each site are distinct, with the ratio at the urban roadside, urban background and rural background sites being 1.6, 1.3 and 0.5, respectively.At the roadside site, the median coarse to fine mass ratio is almost triple that observed at the rural background; this is consistent with the dominant source of PM at the roadside site being the resuspension of large dust particles.At the rural site, the PM size distribution has a greater ratio of fine material consistent with the rural site having a signature of regional background PM.The ratio of coarse to fine PM at the urban background site is intermediate between the roadside and rural background sites, which suggests that this site is significantly effected by both the regional background and the urban road PM sources.These insights into the coarse to fine PM ratio are consistent with the roadside and urban increments, shown in Fig. 5 and discussed previously.

Comparison with previous measurements
To the best of our knowledge, there has been no previous literature study to date using calibrated low-cost sensors to measure PM in Nairobi.Furthermore, it is difficult to make a comparison with previous PM studies in Nairobi because of the differences in the temporal resolution of the data and campaign durations used in this study compared to past measurements.
The most comparable study of PM 2.5 would be the work of Gaita et al. (2016), which also recorded the levels of PM 2.5 at the University of Nairobi (urban background site).The urban background average of PM 2.5 during this study's campaign period was 24.8 µg m −3 compared to the Gaita et al. (2016) mean average of 9.8 µg m −3 , showing a significant increase of 253 %.The sampling time window used in the Gaita et al. (2016) study was between August and September 2007, which is distinct from the February to March 2017 period of this study.Both of these study periods were largely dry, with low precipitation levels, thereby suggesting PM deposition would have been similar between the two studies.The significant increase in measured PM 2.5 could be due to several reasons.Firstly, there could be seasonal differences between August/September and the February/March sampling periods of the two studies; however, the study of Gaita et al. (2014) suggests the urban background concentrations of PM 2.5 mass concentration is similar between these two time periods.The regional background PM loading may have increased during this time period, potentially due to increasing regional aridity caused by climate change leading to more dust generation (Greve et al., 2017).There is almost 10 years difference in time between this study compared to Gaita et al. (2016), in this time Nairobi has undergone significant increases in population and urbanization with correspondingly higher use of motorization and fuel.Using UN population data (UN, 2014), the population of Nairobi is well modelled by Eq. ( 1), in which Y is year date, and p is the population in thousands, which suggests that the population of Nairobi has increased by 148 % from 2007 to 2017.
Hence, the population increase alone cannot account for the increase in PM concentration.The pollution production capability per capita could have increased, which is very likely because of the increased rates of motorization and fuel use.
If we assume that the increase in PM is solely due to population increase and per capita pollution, it suggests that in 2017 the average citizen is 70 % more polluting than the average citizen in 2007.The Egondi et al. (2016) study of PM 2.5 in two slums in Nairobi reported much higher values of 166 and 67 µg m −3 for two different slum areas within Nairobi.These values are much higher than the average PM 2.5 values from this study; Egondi et al. (2016) stated that the reason for such high levels of PM 2.5 stemmed from the local situation and distinct sources of PM within the two slums.This study used a TSI OPC, which was placed 1.5 m a.g.l.Therefore, these measurements were likely highly influenced by resuspended dust.
Although Kinney et al. (2011) measured PM 2.5 levels at four roadside locations, the sampling window was only 11 h and therefore it is not possible to directly compare this study to it.However, considering the diurnal variation in PM found in this study, both investigations measured similar PM 2.5 levels.Kinney et al. (2012) recorded daytime concentration ranges of 10.7 and 98.1 µg m −3 for a rural and urban roadside site, respectively, compared to ca. 25 and ca.150 µg m −3 for this study.Again, the increase between sampling years may be a reflection of the increased population, vehicular traffic and rapid urbanization.

Discussion
In this study, we have shown that Nairobi currently has very high levels of PM mass concentration in the PM 1 , PM 2.5 and PM 10 mass fractions.These measurements were conducted using low-cost calibrated OPC-N2 sensors.The measured PM 2.5 and PM 10 concentrations at both the urban roadside and urban background sites regularly exceeded the WHO daily limits and very likely exceed the annual limits.In particular, the roadside site often showed concentrations that are double the WHO guidelines.These concentrations will very likely be causing significant harm to the population of Nairobi.
The negative health effect of PM is linked to the level of exposure experienced by the patient.This paper and others (e.g.Gaita et al., 2014) have shown that in Nairobi, vehicle emissions are the most significant source of PM.Hence, in Nairobi and other similar cities, the exposure to outdoor PM is to a large extent a function of ones proximity to roads.Furthermore, since traffic varies diurnally, seasonally and by day of the week, personal exposure is both spatially and temporally dependent.This spatial and temporal heterogeneity leads to health inequalities in cities.The urban poor who are often most vulnerable to environmental risks due to lack of adequate health provision, typically live in close proximity to roadways, heightening their exposure to vehicular emissions.Stemming from poorly planned rapid urbanization and inadequate service provision within these cities, those that are unable to afford public transport or personal vehicles frequently walk along these pollution heavy roads, only increasing their exposure periods (Hajat et al., 2015).
This study only looked at outdoor air quality, it is important to stress that a significant amount of air pollution deaths in Kenya and SSA in general are due to poor indoor air quality (Mannucci and Franchini, 2017).As a total number of deaths, deaths related to indoor air quality in Kenya rose to 18 % from 1990 to 2013 (Roy, 2016).In LMICs, indoor exposure to pollutants is typically from the household combustion of solid fuels on open fires or traditional stoves (Yip et al., 2017).These exposures increase the risk of acute lower respiratory infections and associated mortality among young children; indoor air pollution from solid fuel use is also a major risk factor for cardiovascular disease, chronic obstructive pulmonary disease and lung cancer among adults (Muindi et al., 2016).
This study has shown that the low-cost OPC-N2 sensors can be used to generate diurnal PM data sets with good precision and repeatability.As noted in the methodology, it would have been preferable for more cross-calibration periods with the tried and tested gravimetric PM measurement, but resources did not allow this.In addition to more calibration points, the study could have been enhanced by the inclusion of collocated calibration points for the roadside and rural background sites in addition to the urban background site, since the average particle shape, size and density will likely be different between the three sites because of differing PM sources and emission factors.However, it is noted that whilst it is desirable from a purely scientific point of view to have more inter-comparison with reference-grade equipment, every inter-comparison adds significant additional cost to the project both in terms of consumables for the gravimetric analysis (including the cost of analytical-grade filters and accompanying laboratory supplies) and the cost of workers.Many other cities in SSA and other LMICs do not have the resource that Nairobi does in having a gravimetric sampler.These additional costs required for highly accurate scientific results would likely make the low-cost sensors not so very low cost after all, and hence bring into question their unique selling point.
Whilst this paper focused on PM pollution, it is noted that there are serious risks to health not only from exposure to PM but also from exposure to ozone (O 3 ), nitrogen dioxide (NO 2 ) and sulfur dioxide (SO 2 ) (Lelieveld et al., 2015).As with PM, concentrations are often highest in the urban areas of LMICs.Ozone is a major factor in asthma morbidity and mortality, while nitrogen dioxide and sulfur dioxide can also play a role in asthma, bronchial symptoms, lung inflammation and reduced lung function.Good quality measurements of these gas phase pollutants lag behind measurements of PM in Nairobi, other SSA cities and LMIC cities in general.This is due to the high importance of PM as an environmental risk factor but also because of the lack of good quality gas analysers which are affordable and transportable.

Conclusions
Air quality in many LMIC urban centres is often poor and in many cities is getting worse due to the combined pressures of increasing population, increasing urbanization, increasing vehicular traffic and poor vehicle regulation.To be able to manage air pollution, good quality and long-term data sets are required.Unfortunately, in many LMICs the cost of certified high-quality air quality measurements is beyond the financial means of environmental authorities.Low-cost sensors offer the possibility of air quality products at significantly lower cost compared to traditional methods.
This paper used calibrated OPC-N2 devices to measure PM concentrations in Nairobi, Kenya, in the size fractions PM 1 , PM 2.5 and PM 10 .The data required calibration using an established gravimetric 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 options for air quality monitoring.PM was measured in three locations: an urban roadside, urban background and rural background site for a period of approximately 2 months.The data reveals that roadside and urban background locations in Nairobi often exceed the WHO guidelines for daily averaged PM mass concentration in both the PM 2.5 and PM 10 size fractions.Comparison of these data with previous measurements conducted in Nairobi is difficult, but where comparison is possible it appears that air quality has become worse in the last 10 years, which is likely due to increases in population, urbanization and motorization.Changes in industry may also influence the air quality.Comparison of the data from the three sites, following a Lenschow-type approach, allowed for the calculation of representative roadside and urban increments for Nairobi (Lenschow et al., 2001).This increment data can be used in future air quality modelling to assess the likely health impact of PM pollution on Nairobi's population.The combination of the diurnal PM data with local meteorology allows for simple source apportionment of the PM.The diurnal PM concentration tracks the Nairobi rush hours; furthermore, PM peaks when the wind comes from the direction of significant numbers of vehicles such as major roads and a matatu stop.These facts taken together point towards vehicle emissions being the major sources of air pollution in Nairobi, as has been previously observed in studies such as Gaita et al. (2014).The coarse PM fraction increases at the roadside site compared to the urban background site suggesting that non-exhaust vehicle emissions make up a significant amount of the vehicle emissions.
In summary, the low-cost sensors used in this study provided many useful data for assessing air quality in Nairobi at an equipment cost significantly lower than that of traditional instruments.Low-cost sensors have great potential in other country settings and could be used for long-term sampling if the appropriate calibrations are performed.
Data availability.Original research data are available from the authors upon request.
The Supplement related to this article is available online at: https://doi.org/10.5194/acp-18-15403-2018supplementAuthor contributions.FDP and MG devised the study.The data analysis was carried out by DN and RB with guidance from FDP and MG.AP designed the air pollution monitoring boxes.FDP and RB led the manuscript writing with input from all authors.
Competing interests.The authors declare that they have no conflict of interest.

Figure 1 .
Figure 1.Locations of data collection sites and wind rose for the urban background location.(a, c) Locations of Nairobi and Nanyuki in Kenya.Panel (b) shows relative locations of urban background locations (University of Nairobi, American Wing) and urban roadside location (Nairobi fire station).(d) Wind rose generated from data collected at the urban background location during the measurement campaign.

Figure 2 .
Figure 2. Hourly time series data showing PM 10 , PM 2.5 and PM 1 mass concentrations at the three study locations.Red line is the urban roadside, black line is the urban background and blue line is the rural background.Where multiple OPC-N2 devices were measuring in the same location at the same time, the average is provided.The grey shading represents rain events as measured at the urban background location.

Figure 3 .
Figure 3. Box and whisker plots of the hourly averaged PM 2.5 and PM 10 mass concentrations measured at the three sites.

Figure 4 .
Figure 4. Diurnal variation in PM 10 , PM 2.5 and PM 1 mass concentration measured at the three field sites during the whole campaign period.(a) shows the measured concentrations.(b) shows the mass concentrations that have been normalized to the average mass concentration.

Figure 5 .
Figure5.Box and whisker plots of urban and roadside increment of PM 10 , PM 2.5 and PM 1 calculated for Nairobi.Data were taken from the intensive campaign period when the urban background, urban roadside and rural background sites were all measuring simultaneously.Hourly averaged mass concentration data are used.

Figure 6 .
Figure 6.Bivariate (polar) plots of PM in different size fractions at urban background (UB) and urban roadside (UR) sites.Panel descriptions (a) PM 1 UB, (b) PM 2.5 UB, (c) PM 10 UB, (d) PM 1 UR, (e) PM 2.5 UB and (f) PM 10 UR.The PM mass concentration data in each plot are normalized to allow for easy comparison between the different sites and PM size fractions investigated.However, note the scale bars are different for each panel to allow for easier interpretation.

Figure 7 .
Figure 7. Ratio of the coarse (PM 10 -PM 2.5 ) to fine (PM 2.5 ) fraction of PM.

Table 1 .
Mean average PM mass concentrations (PM 1 , PM 2.5 and PM 10 ) and daily exceedances of the WHO PM guidelines (PM 2.5 and PM 10 ) observed at the three measurement sites during the campaign period.WHO guidelines for daily PM 10 and PM 2.5 are 50 and 25 µg m −3 , respectively.

Table 2 .
Summary of roadside and urban increments for the PM 1 , PM 2.5 and PM 10 size fractions measured during the intensive period.Roadside increment (µg m −3 ) Urban increment (µg m −3 )