Spatial-temporal patterns of ambient fine particulate matter (PM_2.5) and black carbon (BC) pollution in Accra

The GAMA is the industrial and administrative center of Ghana and one of the fastest growing metropolitan areas in SSA with ∼5 million residents and an annual growth rate of 4.2% [24]. The GAMA consists of Accra Metropolitan Area (AMA) at its core, the port city of Tema to the east and 11 other adjoining districts [25, 26]. The GAMA is in a tropical climate zone with high average monthly temperatures and relative humidity (RH) ranging between 25 °C and 33 °C (77–90 °F) and 77%–85%, respectively [25]. The GAMA has two major seasons: the rainy (May–October) period, and the dry period comprising the Harmattan (November–February) characterized by north-easterly trade winds from the Sahara Desert [27].

This work was conducted within the multi-country and multi-city 'Pathways to Equitable Healthy Cities' study (http://equitablehealthycities.org/), which aims to provide scientific evidence on how urban development and policies can be managed to enhance health equity.

As previously described [27], we designed a year-long campaign to examine the spatial (land-use features) and temporal (daily, weekly, monthly and seasonal) variations in ambient PM_2.5 and BC by sampling at a combination of fixed (∼1 year, n = 10 sites) and rotating (7 d, n = 136) sites. This design allowed for detailed assessment of both the temporal (using fixed site data) and spatial (using rotating site data) variability of PM_2.5 and BC over the study area. Further, this design allowed us to optimally use a finite number of monitoring equipment to capture data across the entire geographical extent of the study area. We used a structured form to collect information on land-use features at each monitoring site [27]. The sites were subsequently grouped into four land-use classes: commercial, business, industrial (CBI); high-density residential; medium/low-density residential; or peri-urban (see supplementary text S1 for additional details). We originally planned a 12 month field campaign to collect data at 150 sites starting April 2019; however, the fieldwork was suspended for six weeks (31st March–18th May 2020) due to the COVID-19 pandemic lockdown in Accra and self-isolation of field team members. After the lockdown was lifted and daily activities returned to pre-lockdown status, we conducted additional three weeks of measurement (19th May–11th June 2020) at all fixed sites along with 12 rotating sites, resulting in close to 12 months of data from 10 fixed and 136 rotating sites (see figure S2 (available online at stacks.iop.org/ERL/16/074013/mmedia) for measurement timeline).

The 10 fixed sites were operated continuously, collecting weekly and 1 min averages throughout the measurement campaign at key locations selected based on population density, road networks, neighborhood socioeconomic status (SES) and household biomass fuel use data from the national census [28]. To compare changes in annual mean PM_2.5 levels within the last decade, four of the 10 fixed sites were placed at the exact locations monitored by Dionisio and colleagues [23]. We also collected 1-week samples at each of the 136 rotating sites, which were selected with a stratified random sampling scheme based on land-use, with more emphasis placed on AMA where the majority of the population live [27]. The sites were initially computer-generated and the actual sampling locations that were as close as possible to the computer-generated ones were identified by the field team. The median distance (interquartile range, IQR) between the original computer-generated locations versus the actual sites monitored was 181 (67–407) m. During the field campaign, the rotating sites were sampled in groups of five each measurement week alongside the fixed sites.

We measured both real-time (1 min interval) and integrated gravimetric (weekly averages) PM_2.5 concentrations using portable battery operated low-cost and low-power monitors that were placed in protective cases fastened on metal poles at about 4 m (±1 m) above ground [27]. We included in our analysis only samples from monitors that operated for ⩾75% of the measurement period (i.e. at least 5 out of 7 d to capture both weekdays and weekends) and had an average flow rate within 10% of the intended rate.

Weekly integrated PM_2.5 was measured using the Ultrasonic Personal Aerosol Sampler (UPAS) (Access Sensor Technologies, Fort Collins, USA) [29] operated at 1 litre per minute (lpm). The UPAS has been demonstrated to have a close agreement with reference monitors [29–31] over a wide range of concentrations (10–1600 μg m⁻³) in diverse settings. However, a recent field evaluation suggested that overloading could occur at filter masses above 650 μg [32], an issue that could be avoided by using the duty-cycle feature on the UPAS in highly polluted environments. To avoid overloading filters and to also conserve battery power, the UPAS was operated at 50% duty cycle, drawing air 30 s every minute for a total of 5040 min over the 7 d sampling period. PM_2.5 mass was collected on 2 µm pore size 37 mm barcoded Teflon membrane filters (https://mtlcorp.com/filters/) and weighed pre- and post-sampling using a MTL AH500 automated robotic scale (www.mtlcorp.com/#/filter-weighing) maintained in a temperature and RH controlled laboratory (23 ± 2 °C, 35 ± 2% RH) at The University of British Columbia. Further information on the UPAS and filter handling can be found elsewhere [27, 33]. An additional 27 duplicate (20% of sites) integrated samples and 28 field blanks were collected at rotating sites, including three post-COVID-19 lockdown duplicates and blanks. The average of the duplicate measurements was taken and final PM_2.5 concentrations were blank corrected. Quantitative information on blanks and duplicates are in the supplementary text (figure S3).

We deployed a low-cost Zefan real-time continuous monitor (www.zfznkj.com/) to measure PM_2.5 concentrations at 1 min intervals. The Zefan relies on a light scattering technique to assess PM_2.5 using Plantower sensors (model PMS7003), which have been evaluated with reference monitors (i.e. FDMS 8500 and TEOM 1400ab) over 6–12 month periods [34, 35]. While this technique provides accurate temporal pattern in measured PM concentrations, its magnitude is inexact as PM mass are only inferred from particle characteristics (e.g. number, size and refractive index), which can be affected by weather conditions (e.g. RH and temperature) [34, 36].

Following previous studies [23, 37], we corrected the minute-by-minute continuous PM measurements by a correction factor (CF) calculated such that the average of continuous PM_2.5 measurements was equal to the integrated gravimetric PM_2.5 concentration at the same location over the same 7 d measurement period. This was done to ensure that the average weekly continuous measurements were the same as the gravimetric which has less error than optical sensors. We calculated unique CFs per site for each 7 d period. The median (IQR) of the CFs were 0.84 (0.69–1.13), similar to CFs previously reported for a different optical sensor in Accra [23, 37].

We tested minute-by-minute monitor-to-monitor precision by running all monitors alongside each other over a 24 h period prior to the commencement of field campaign [27]. Further, we conducted mid-campaign (in January 2020) monitor-monitor precision by co-locating the instruments at one of the fixed sites for a week to assess potential drift over the course of the campaign. Finally, post-campaign, we co-located two Zefan sensors with a U.S. federal reference monitor located at the U.S. embassy in Accra. We did not see any within- or between-monitor bias in the sensor performance pre-, mid-, and post-campaign.

Black carbon (BC) aerosols are known indicators of combustion-related constituents of PM emissions and contribute to global warming [38, 39]. Recent epidemiological studies also indicate associations between BC and adverse health outcomes [16, 40]. Thus, we used the absorption coefficient (light absorbance) (10⁻⁵m⁻¹) of the post-weighed PM_2.5 filters, estimated by applying an image-based reflectance method [41], as a marker for BC concentrations [42, 43]. The image-based reflectance method closely correlates (r² = 0.98) to elemental carbon (EC) concentrations by thermo-optical reflectance, with 1 absorbance unit (1 × 10⁻⁵m⁻¹) equivalent to 1.67 µgm⁻³ EC [41].

We used data from the rotating sites to assess the spatial patterns of PM_2.5 and BC across the city by the four site-types: CBI, high-, and medium/low-density residential and peri-urban. To provide more detail on influence of traffic related sources on PM_2.5 pollution in the GAMA, we grouped the samples collected at rotating sites according to the type (major, secondary and minor) and surface material (paved, mixed and unpaved) of the road near the monitoring site. Since monitoring at rotating sites occurred in groups of five sites per week (i.e. samples were not collected simultaneously at all rotating sites, nor evenly by site-types during each measurement week), we accounted for potential influence of time trend/season on the spatial patterns of the measured concentrations to allow for comparison across sites. We adjusted for potential time trends at the rotating sites by applying weekly specific temporal adjustment factor (TAF) using data from the ten fixed (year-long) sites. For each measurement week, a TAF calculated as the ratio of the mean PM_2.5 or BC across all fixed sites for that week to the mean annual PM_2.5 or BC across all fixed sites was used to adjust the samples collected at the rotating sites in that particular week [44]. The season adjusted concentration $({C_i})_{j{ }}^{{\text{adjusted}}}$ of the ith rotating site for the jth measurement week was calculated as:

$$\begin{equation}({C_i})_j^{{\text{adjusted}}} = {({C_i})_j}/\left[ \left( C^{{\text{Fixed Site}}}\right)_j/\left( {\overline {C^{{\text{Fixed Site}}}} } \right) \right]\end{equation} \tag{ 1 }$$

where ${({C_i})_j}\,$is the PM_2.5 or BC concentration measured at the ith rotating site in the jth measurement week; ${(C^{{\text{Fixed Site}}})_j}$ and $\left( {\overline {C^{{\text{Fixed Site}}}} } \right)$ are the average PM_2.5 or BC in the corresponding jth measurement week and annual average PM_2.5 or BC at all fixed sites respectively, and $\left[ \left( C^{{\text{Fixed Site}}}\right)_j/\left( {\overline {C^{{\text{Fixed Site}}}} } \right) \right]$ is the TAF.

We examined the temporal patterns in the data by season (Harmattan vs non-Harmattan), days of the week (plus weekday vs weekend), and time of day (diurnal) using data from the fixed sites. We also evaluated changes in annual PM_2.5 levels over a decade (2006–2007 vs 2019–2020) by comparing fixed site data obtained from the same four residential locations sampled in a previous study [23].

All analyses were done using the statistical analysis package R, version 3.6.1 [45], and an alpha of 0.05 was used as cut-off of significance.

The measurement locations and the measured concentrations relative to the World Health Organization (WHO) air quality guideline are shown in figure 1. The season adjusted mean (standard deviation, SD) integrated PM_2.5 and BC concentrations across the rotating sites were 31 (10) μg m⁻³ and 5 (2) × 10⁻⁵m⁻¹ respectively. PM_2.5 concentration at every rotating site was higher than the WHO annual guideline of 10 ug m⁻³, while 99%, 71% and 31% of the sites exceeded the interim target 3 (IT-3, 15 μg m⁻³), IT-2 (25 μg m⁻³) and IT-1 (35 μg m⁻³), respectively (figure 1). The mean PM_2.5 and BC levels at rotating sites varied by land-use. The highest PM_2.5 concentrations were in CBI areas (mean: 37; range: 23–67 μg m⁻³) and high-density residential neighborhoods (mean: 36, range: 21–67 μg m⁻³) (p < 0.01). Peri-urban sites had the lowest concentrations (mean: 26, range: 16–56 μg m⁻³) after medium/low-density neighborhoods (mean: 28, range: 15–54 μg m⁻³) (figure 2). Similarly, BC concentrations were two times higher in CBI areas (mean: 7, range: 1–14 × 10⁻⁵m⁻¹) compared with peri-urban sites (mean: 3, range: 1–6 × 10⁻⁵m⁻¹) (table 1). In general, average PM_2.5 concentrations were slightly higher at sites along major and secondary roads compared with sites near minor roads, but not by road surface. We observed similar patterns for BC. Overall, the relative differences in BC across land use factors were much larger than the relative differences in PM_2.5 concentrations, suggesting that PM_2.5 in the GAMA may not be affected by community/local sources (such as vehicle tailpipe emissions and trash burning) as much as BC.

Zoom In Zoom Out Reset image size

Zoom In Zoom Out Reset image size

Mean (SD) annual PM_2.5 concentrations across the ten year-long (fixed) sites was 37 (40) μg m⁻³ and ranged from site-type specific annual means of 26 μg m⁻³ at the peri-urban site, 32–40 μg m⁻³ at medium/low-density residential sites, 35–40 μg m⁻³ at high-density residential sites, and 37–43 μg m⁻³ at CBI areas (figure 3). Similarly, annual mean BC concentrations were lowest at the peri-urban site and highest at CBI sites (table 2).

Zoom In Zoom Out Reset image size

By season, the mean PM_2.5 and BC concentrations during the Harmattan (89 μg m⁻³ and 12 × 10⁻⁵m⁻¹) were 4- and 2-fold higher than the non-Harmattan period (23 μg m⁻³ and 6 × 10⁻⁵m⁻¹), respectively (figure 4). The absolute mean difference in PM_2.5 concentrations between the Harmattan and non-Harmattan periods at each site were between 56 and 71 μg m⁻³ (figure 3). While the absolute levels were higher during the Harmattan, both periods showed substantial relative spatial variability. The peri-urban site recorded the highest seasonal mean difference in PM_2.5 concentrations while sites in high-density residential neighborhoods recorded the lowest. For each measurement month, the peri-urban site consistently registered the lowest PM_2.5 and BC levels (figure 4). Like PM_2.5, BC levels also increased during the Harmattan months (figure 4(b)), and both showed higher variability in the Harmattan as indicated by the sample SD. The overall observed doubling of BC levels in the Harmattan period is noteworthy as it indicates that meteorological conditions likely magnify local emissions and lead to higher concentrations.

Zoom In Zoom Out Reset image size

Using the minute-by-minute continuous data, we found no differences in mean PM_2.5 concentrations between day of the week (Monday–Sunday) nor between weekdays and weekends in the GAMA, regardless of whether the data were from the fixed or rotating sites or both (p > 0.05). Although Sundays showed slightly lower mean PM_2.5 overall, the mean difference (3 μg m⁻³) was not significant (p = 0.57). The absence of between-day of the week variation in PM_2.5 in the GAMA was consistent across all land-use categories (figure S4).

PM_2.5 concentrations from all sites showed strong bimodal variability across time of day, and was consistent over land-use areas and by season (figure 5). PM_2.5 concentrations at all sites rose around 03:00 daily, peaking at about 06:00, followed by a gradual decline to their lowest values around 10:00. Levels remained fairly stable between 10:00 and 15:00, after which the concentration slowly increased with a relatively smaller peak around 18:00–19:00. There was about an hour delay in the timing of the peaks during the Harmattan and the smaller early evening peak was less pronounced compared to the non-Harmattan period. In general, average PM_2.5 concentrations at nighttime (18:00–05:59) were slightly higher than daytime levels (37 vs 34 μg m⁻³). During these periods, biomass is burned in some neighborhoods for residential and small-scale commercial purposes, such as cooking street food and bakery operation.

Zoom In Zoom Out Reset image size

In 2006/2007, Dionisio and colleagues [23] recorded large variability (with wide SDs) in mean annual PM_2.5 in four residential neighborhoods of varying SES and biomass use within the AMA, with values ranging from 28 μg m⁻³ in the affluent neighborhood of East Legon (EL), and 57 μg m⁻³ in middle-income Asylum Down (AD), to >70 μg m⁻³ in low-income, densely populated Nima (NM) and Jamestown (JT) (figure 6). In the current study (2019/2020), the mean annual PM_2.5 concentrations were lower at the same locations, and ranged from 34 μg m⁻³ at EL to 40 μg m⁻³ in JT. This suggests a reduction (and more uniformity/plateau) in PM pollution in the city. The largest reductions were observed in high-density residential neighborhoods of JT and NM, where PM_2.5 levels decreased on average by ∼60%. We observed a smaller reduction (35%) in the middle-income AD, but slight increase (21%) in high-income EL where there are a mix of residences and corporate, commercial and small businesses. The observed increase in high-income EL could also come from an overall increase in local commercial activities.

Zoom In Zoom Out Reset image size