Celeste McFarlane1,2, Paulson Kasereka Isevulambire3, Raymond Sinsi Lumbuenamo4, Arnold Murphy Elouma Ndinga5, Ranil Dhammapala6, Xiaomeng Jin7, V. Faye McNeill2,8, Carl Malings9,10, R. Subramanian9,11,12, Daniel M. Westervelt This email address is being protected from spambots. You need JavaScript enabled to view it.1,13

1 Lamont-Doherty Earth Observatory of Columbia University, New York, USA
2 Columbia University, Department of Chemical Engineering, New York, USA
3 Ecole Régionale postuniversitaire d’Aménagement et de Gestion Intégrés des Forêts et Territoire tropicaux (ERAIFT) Kinshasa, Democratic Republic of Congo
4 World Bank Group, Kinshasa, Democratic Republic of Congo
5 Département de chimie, Université Marien Ngouabi, Brazzaville, Republic of Congo
6 Washington State Department of Ecology, Washington, USA
7 Department of Chemistry, University of California Berkeley, USA
8 Columbia University, Department of Earth and Environmental Sciences, New York, USA
9 OSU-EFLUVE - Observatoire Sciences de l’Univers-Enveloppes Fluides de la Ville à l’Exobiologie, Université Paris-Est-Créteil, France
10 NASA Postdoctoral Program Fellow, Goddard Space Flight Center, Greenbelt, Maryland, USA
11 Center for Atmospheric Particle Studies, Carnegie Mellon University, Pittsburgh, PA 15218, USA
12 Kigali Collaborative Research Centre, Kigali, Rwanda
13 NASA Goddard Institute for Space Studies, New York, USA

Received: November 4, 2020
Revised: February 6, 2021
Accepted: March 22, 2021

 Copyright The Author(s). This is an open access article distributed under the terms of the Creative Commons Attribution License (CC BY 4.0), which permits unrestricted use, distribution, and reproduction in any medium, provided the original author and source are cited.

Download Citation: ||https://doi.org/10.4209/aaqr.200619  

Cite this article:

McFarlane, C., Isevulambire, P.K., Lumbuenamo, R.S., Ndinga, A.M.E., Dhammapala, R., Jin, X., McNeill, V.F., Malings, C., Subramanian, R., Westervelt, D.M. (2021). First Measurements of Ambient PM2.5 in Kinshasa, Democratic Republic of Congo and Brazzaville, Republic of Congo Using Field-calibrated Low-cost Sensors. Aerosol Air Qual. Res. 21, 200619. https://doi.org/10.4209/aaqr.200619


  • We developed multiple linear regression applied to PM2.5 data in Kampala, Uganda.
  • Mean absolute error reduced from 14.8 µg m3 in the raw PM2.5 dataset to 3.4 µg m3.
  • The first ever measurements of ambient PM2.5 are presented in Congo.
  • Annual mean PM2.5 is 4 times the WHO Interim Target 1.
  • Surface PM2.5 and AOD were each about 40% lower in the region COVID19.


Estimates of air pollution mortality in sub-Saharan Africa are limited by a lack of surface observations of fine particulate matter (PM2.5). Despite being large metropolises, Kinshasa, Democratic Republic of the Congo (DRC), and Brazzaville, Republic of the Congo (ROC), which possess populations of 14.3 million and 2.4 million, respectively, use no reference air pollution monitors at the time of writing. However, a few reference monitors have recently been deployed in other parts of sub-Saharan Africa, including a Met One Beta Attenuation Monitor (BAM-1020) at the U.S. Embassy in Kampala, Uganda, next to which a low-cost PM2.5 monitor, the PurpleAir, was collocated in August 2019. The raw PurpleAir data from September 2019 through February 2020 strongly correlated with the BAM-1020 measurements (R2 = 0.88) but also exhibited a mean absolute error (MAE) of approximately 14 µg m–3. Employing two calibration models, namely, multiple linear regression and random forests, decreased the MAE to 3.4 µg m–3 and increased R2 to 0.96. Given the similarity in climate and emissions, we applied the collocated field correction factors for Kampala to four PurpleAir units in Kinshasa and one in neighboring Brazzaville, which were deployed in April 2018. We estimated an average annual PM2.5 concentration of 43.5 µg m–3 in Kinshasa for 2019, which exceeds the World Health Organization (WHO)’s Interim Target 1 (10 µg m–3) by 4 times. Finally, the surface PM2.5 level and the aerosol optical depth were about 40% lower during the COVID-19 lockdown in 2020 than the corresponding period in 2019, which cannot be attributed solely to changes in meteorology or wildfire emission. Hence, our results highlight the need to implement clean air solutions in the Congo.

Keywords: Low-cost sensors, Particulate matter, Air quality, Africa


Ambient air pollution is a major global public health crisis that causes an estimated 4.9 million premature deaths per year around the world. Air pollution is the fifth leading risk factor for all mortality (Health Effects Institute, 2019), and on average reduces life span by 20 months worldwide, rivaling the global impact of cigarette smoking. According to one air quality modeling study, air pollution may cause at least an estimated 780,000 premature deaths annually in Africa (Bauer et al., 2019), and a significant number of diseases (co-morbidities) are known to be worsened by chronic exposure to air pollution, like asthma, lung cancer, and chronic obstructive pulmonary disease (Burnett et al., 2018). However, sparse air pollution monitoring imparts high uncertainty to estimates of exposure and impact. Most of these ambient air pollution deaths come from PM2.5, or particulate mass concentrations for particles with diameter less than 2.5 µm.

There is currently no publicly available PM2.5 monitoring in the city of Kinshasa, Democratic Republic of Congo, or Brazzaville, Republic of Congo, either as regulatory monitors or citizen-deployed low-cost sensors. Therefore, it is currently difficult to know the level of exposure and the potential health impacts of air pollution in Kinshasa and Brazzaville, an alarming gap in a pair of capital megacities containing more than 16 million people and experiencing rapid population growth. To address the lack of data in both cities, we deployed and calibrated a small network of 5 low-cost air quality sensors starting in 2018.

Almost no prior published work has considered PM2.5 or air quality in Kinshasa or Brazzaville to this point. Mbelambela et al. (2017) used a portable sensor to calculate personal PM2.5, NO2, and SO2 exposure in a cohort of 517 subjects, mostly bus drivers. Data were only collected between 20 April and 14 May of 2015 and are more representative of personal exposure rather than ambient concentrations. The authors reported PM2.5 exposure over the 3-week period ranging from 64 to 129 µg m–3. The only other air quality study in the peer-reviewed literature in either city was a study of trace metals conducted in 1990 (Lobo et al., 1990), before the Democratic Republic of the Congo was known as such, and has limited relevance to present-day exposures. No studies on air quality in neighboring Brazzaville were found in the literature. Additionally, to our knowledge, there are no national ambient air quality standards in either country.

Low-cost sensors (LCSs) have the potential to improve air quality data coverage throughout the world, especially in resource-limited areas (Amegah, 2018). For LCSs to provide high-quality data, understanding local conditions is vital. Calibration factors and sensor technical performance vary strongly with conditions such as temperature, relative humidity, particle size distribution, and particle loading (Levy Zamora et al., 2019; Hagan and Kroll, 2020; Tryner et al., 2020). Other factors such as sensor aging can also influence sensor performance and have been studied in both a laboratory and field setting (Malings et al., 2019; Tryner et al., 2020). Careful collocation, or side-by-side placement of LCSs with reference monitors, is an essential step to getting accurate data out of LCSs. Several recent studies have performed either a field calibration or a laboratory calibration for a variety of different sensors in a variety of different environmental conditions, paving the way for additional studies in diverse environments (Holstius et al., 2014; Jiao et al., 2016; Johnson et al., 2016; Kelly et al., 2017; Jayaratne et al., 2018; Tryner et al., 2019; Hagan and Kroll, 2020; Tryner et al., 2020). However, to our knowledge, published LCS collocation and performance evaluation studies within sub-Saharan Africa are limited (Subramanian et al., 2020).

Here we present the first-ever multi-year, field-calibrated ambient PM2.5 dataset in Kinshasa and Brazzaville. We first build a multiple linear regression (MLR) model based on a collocation of sensors and reference (Federal Equivalence Method (FEM)) PM2.5 monitors in Kampala, Uganda, and develop a correction factor for low-cost sensors. We then apply the correction to our network in Kinshasa and Brazzaville. We analyze PM2.5 on monthly, weekly, daily, and hourly timescales. We interpret the data in the context of changing meteorology and changing human activity coinciding with COVID-19-related stay-at-home orders. Finally, we assess the air quality picture in Kinshasa as seen from satellite remote sensing.


2.1 Description of PurpleAir Low-cost Sensors

We deploy five PurpleAir PA-II-SD low-cost PM2.5 devices (www.purpleair.com) in the Kinshasa-Brazzaville area and one PurpleAir device in Kampala for calibration purposes. PurpleAir uses dual Plantower PMS 5003 optical sensors to estimate PM2.5 mass concentrations and a Bosch BME280 sensor to estimate temperature, relative humidity, and pressure. Data are transmitted via wireless connectivity in real time and recorded to an on-board 16 GB microSD card. All PurpleAir data is available online on their website. The devices measure sensor readings in six size bins ranging from 300 nm to 10 µm at approximately a 60-s interval. A proprietary algorithm converts raw sensor measurements to PM1, PM2.5, and PM10 mass using assumptions about particle shape and density. We use the CF=ATM data field as provided by PurpleAir, which is actually the higher CF=1 data field on account of mislabeled columns by PurpleAir. The CF=1 data has not been transformed nonlinearly and is a better input into regression models. PurpleAir PM2.5 is known to strongly correlate (R > 0.9) with reference-grade monitors but are subject to biases at high relative humidity in particular (Jayaratne et al., 2018; Malings et al., 2019; Magi et al., 2020; Tryner et al., 2020). Sensitivity to relative humidity has also been identified in other low-cost air pollution monitoring devices (Di Antonio et al., 2018; Jayaratne et al., 2018; Kelleher et al., 2018; Hagan and Kroll, 2020; He et al., 2020). PA-II sensors cost approximately US$250 per unit which are about 100 times cheaper than reference PM2.5 monitors, making them attractive for multi-sensor networks. Their use, however, requires careful field calibration in order to achieve high-quality data. Malings et al. (2019) developed a multiple linear regression-based calibration method which will be utilized in the following sections.

2.2 Sampling Locations and Periods

2.2.1 Kampala

A PurpleAir was collocated at the U.S. Embassy in Kampala, located at Plot 1577 Ggaba Road, about 3 km from the city center, 0.301268 N latitude and 32.591711 E longitude. Sampling began in September 2019 and continues through the present. Also located at the U.S. Embassy is a Met One Beta Attenuation Monitor (BAM-1020), which provides a calibration point for the PurpleAir. Data collection of PM2.5 with the BAM-1020 has been ongoing at the U.S. Embassy since January 2017 (https://www.airnow.gov/international/us-embassies-and-consulates/, last accessed 26 August 2020). We use the September 2019–February 2020 overlapping dataset with the BAM-1020 and PurpleAir for our calibration (see “Field calibration” in “Methods”).

2.2.2 Kinshasa and Brazzaville

Four PurpleAir sensors were deployed throughout Kinshasa with a fifth in Brazzaville. Starting in March 2018 and continuing through present, a device was located at the U.S. Embassy in Kinshasa at 310 Avenue des Aviateurs, latitude 4.3002 S and longitude 15.3138 E. In November of 2019, sensors were added at the Université Pédagogique Nationale (UPN) located on Route de Matadi, latitude 4.4039 S and longitude 15.2572 E, and the Ecole Régionale Postuniversitaire d’Aménagement et de Gestion intégrés des Forêts et Territoires tropicaux (ERAIFT), latitude 4.4103 S and 15.3065 E. A fourth sensor was located in a residential area in the Kintambo area of Kinshasa at the Belle Vue Villas (CBV), latitude 4.3278 S and longitude 15.2722 E, starting in November 2019. Finally, a fifth sensor was located across the Congo River in Brazzaville, Republic of Congo at the U.S. Embassy at latitude 4.2751 S and longitude 15.2561 E, starting in February 2020. A map of all sites is shown in Fig. 1.

Fig. 1. Map of sensor locations in Kinshasa and Brazzaville. Background map © Google, 2020.Fig. 1. Map of sensor locations in Kinshasa and Brazzaville. Background map © Google, 2020.

2.3 Field Calibration

2.3.1 Multiple linear regression

We use both an MLR and a random forest approach for bias-correcting the PurpleAir data towards the reference monitor in Kampala. The collocation of the PurpleAir and the FEM monitor took place between September 2019 and March 2020. A randomly selected 75% of data during that time period is used to build the multiple linear regression model, and the remaining 25% is used for validation. This model was developed using the base R statistics package. The multiple linear regression approach follows a similar methodology as in Malings et al. (2019) in which daily averaged raw PurpleAir PM2.5, relative humidity, and temperature are used as explanatory variables to predict corrected PM2.5 concentration:


The multiple linear regression model was evaluated based on coefficients of determination (r2) and MAE defined as:


where PM refers to PM2.5 (in µg m–3), ref is the BAM-1020 reference monitor, LCS refers to the low-cost sensor data, N is total number of observations, and i is the time series variable. Since the MLR and RF methods result in similar correlation and mean bias improvements when the same averaging time periods are used and the MLR approach is more transparent (see “Results and Discussion”), we present further calibrated PM2.5 results based on MLR. Results obtained using the RF approach are not substantially different so they are presented in the supplementary information. The results of the MLR model correction are shown in the “Results and Discussion” subsection “Kampala co-location analysis.”

2.4 Satellite Data

We use remote sensing data from the NASA Aqua and Terra satellites, which contain the Moderate Resolution Imaging Spectroradiometer (MODIS). MODIS views the Earth surface over the Congo region once per day with a daily overpass time around 12:30. We specifically use the aerosol optical depth (AOD) at 550 nm wavelength using the Multi-Angle Implementation of Atmospheric Correction (MAIAC) Level 2 gridded data over land surfaces at 1 km pixel resolution (MCD19A2) (Lyapustin et al., 2018). We convert the pixel data in the H19V9 tile to latitude and longitude points using the MODLAND Tile Calculator (https://landweb.modaps.eosdis.nasa.gov/cgi-bin/developer/tilemap.cgi, last accessed 25 Aug 2020). Cloudy pixel data are removed as part of this data product, and only “best-quality” flagged data are used in our analysis. Data are available on the NASA Earth data repository (https://earthdata.nasa.gov/, last accessed 25 Aug 2020). We use daily MAIAC data from January 2018 through July 2020 in our analysis.

2.5 Meteorological Data

We use the NOAA National Centers for Environmental Information land-based station data for meteorology. In Kinshasa and Brazzaville, the most complete temperature records are at the N’Djili International Airport (Kinshasa) and Maya-Maya International Airport (Brazzaville). Temperature, relative humidity, and wind speed are mostly a complete record at these locations. Greater than 58% of precipitation data between September 2017 and August 2020 are missing at each of these stations, rendering any precipitation analysis impossible. We therefore use temperature, relative humidity, and wind speed data from N’Djili Airport to analyze the impact of weather on PM2.5 observations.


3.1 Kampala Co-location Analysis

In Table 1 we present the regression parameters for the Kampala co-location. Correlation between raw PurpleAir data and Embassy BAM-1020 is high (r2 = 0.88). The overall MAE in the raw data compared to the embassy BAM is 14.8 µg m–3 which is partially driven by a large overprediction during times of very poor air quality (values greater than 100 µg m–3). Values less than 100 µg m–3 show less bias compared to the BAM-1020. Using the MLR model to correct the raw PM2.5 data towards FEM standard using the BAM-1020 results in a reduction of MAE to 3.4 µg m–3, quantified using only the remaining 25% of the collocated data reserved for testing and validation. The corrected PurpleAir PM2.5 data (turquoise line in Fig. 2) shows a substantial improvement in comparison against BAM PM2.5. Correlation is also improved by the MLR model (r2 = 0.96). Though hourly averaged PM2.5 data from the PurpleAir and the BAM-1020 are noisy and therefore not as well suited for MLR, we also build an MLR model for the hourly averaged PM2.5 data, which we only apply to any analysis that uses hourly mean data instead of daily mean data. Table 2 also contains the hourly based MLR statistics.

Table 1. Regression coefficients of the MLR model.

Fig. 2. Performance evaluation and calibration of PurpleAir PM2.5 versus Federal Equivalent Method (FEM) PM2.5 between September 2019 and February 2020 at the U.S. Embassy in Kampala, Uganda. Raw daily data is shown in purple, FEM data in orange, and corrected low-cost sensor data (using the MLR method) in turquoise.Fig. 2. Performance evaluation and calibration of PurpleAir PM2.5 versus Federal Equivalent Method (FEM) PM2.5 between September 2019 and February 2020 at the U.S. Embassy in Kampala, Uganda. Raw daily data is shown in purple, FEM data in orange, and corrected low-cost sensor data (using the MLR method) in turquoise.

 Table 2. Statistics of the MLR and RF models.

We apply the correction factors developed from the long-term MLR collocation in Kampala to each of the five PurpleAir PM2.5 monitors deployed in Kinshasa and Brazzaville. Kampala is just under 2000 km away from Kinshasa; however, this is the closest reference PM2.5 monitor to Kinshasa. The climates of Kinshasa and Kampala are quite similar as both lie within the center of the tropics. Both cities have two wet seasons that peak around October–November and March–April. Unlike Kampala, Kinshasa has a true dry season in which there is no or very little precipitation. Kampala’s “drier” season coincides with Kinshasa’s dry season (June–August). Annual mean temperature and relative humidity are 30.4°C and 80% in Kinshasa and 27.8°C and 75% in Kampala. Differences in emissions characteristics in the two cities are also a possible influence on applicability of our calibration. Thus, we also analyzed emissions data from the Diffuse and Inefficient Combustion Emissions Inventory in Africa (DICE-Africa), comparing source profiles for SO2, black carbon (BC), and organic carbon (OC) emissions in a 25 km × 25 km grid boundary in each city. Results (Fig. S1) indicate that the source profiles have many similarities in the two cities, in particular household fuel usage, which comprises about 50–75% of the total emissions of SO2, BC, and OC in each city. Though the total emissions are slightly higher in Kinshasa compared to Kampala, the similar source mix further justifies applicability of sensor calibrations between the two cities. Qualitative information about particle size in the two cities can be obtained through satellite retrievals of the Angstrom exponent, which we present in Fig. S2. The 2019 annual mean Angstrom exponent as retrieved by MODIS Terra Deep Blue retrieval algorithm is about 1.5 in Kampala and 1.6 in Kinshasa, indicating qualitative similarity in particle size distributions in each city. Though the geographic distance between Kinshasa and Kampala is a limitation of the work, given the similarities between the two environments, estimates of emissions sources, qualitative similarity in particle size, and the paucity of reference monitoring and resources for such monitoring in sub-Saharan Africa, we consider our sub-continental calibration to be sufficient for further analysis.

3.2 PM2.5 Time Series at Each Sampling Location

Fig. 3 shows both the corrected (turquoise) and raw (purple) monthly averaged PurpleAir PM2.5 data collected at the U.S. Embassy in Kinshasa (see “Methods” and Fig. 1) from March 2018 through July 2020. Often the calibrated data and raw data are within 10 µg m–3 of each other, except when the PM2.5 is above 100 µg m–3 or below 25 µg m–3. The 95% confidence interval for daily averaged raw data ranges between ±0.1 to ±0.5 µg m–3, depending on the exact month in the data. PM2.5 data are nearly always above WHO air quality guidelines (10 µg m–3 on an annual mean basis, 25 µg m–3 on a 24-h mean basis). The 2018 average PM2.5 was 54.4 µg m–3, which was followed by a decrease in 2019 (43.5 µg m–3). Through July 2020 (time of last data collection), the average is 35.7 µg m–3, though PM2.5 is typically high in August and later in the year. The 2018 average also does not include the months of January and February, which are typically lower in PM2.5. In addition to this, the decrease in PM2.5 between 2018 and 2019 is not expected to be related to emissions reductions. Emissions trends in recent years for DRC are not well known. However, to our knowledge there has been no emissions control mitigation or measures in the last 3 years. Meteorological data are also limited in Kinshasa and Brazzaville. In Fig. 3 we also plot the Kinshasa station’s available meteorological variables from January 2018 to August 2020. To first order, several previous studies have reported that on a large scale, PM2.5 can be positively correlated with temperature, negatively correlated with wind speed, and positively correlated with relative humidity (Jacob and Winner, 2009; Fiore et al., 2015; Westervelt et al., 2016). As seen in Fig. 3, dry season July 2019 temperatures are approximately the same between 2018, 2019, and 2020, suggesting that the temperature effects on PM2.5 cannot explain the observed decrease in PM2.5 in 2020. Relative humidity in 2020 is slightly higher or roughly the same compared to either 2019 or 2018, inconsistent with the strong 2020 PM2.5 decreases. Wind speed is slightly lower in 2020 compared to 2018 or 2019, which would also have the effect of likely increasing PM2.5, and therefore cannot explain the observed decreases in PM2.5. We conclude that the decrease in PM2.5 between 2018, 2019, and 2020 cannot be directly attributed to at least these three meteorological factors. Another plausible explanation is a potential decrease in wildfire activity in the Congolese rainforest. Burned area was shown to decline by ~1.3% y–1 between 2003 and 2017 in the Central African Republic and South Sudan (Jiang et al., 2020), thousands of kilometers from Kinshasa. Closer to Kinshasa, burned area was found to largely remain unchanged over the same time period. Given these contrasting results, it is not likely that fire activity is the main driver of observed PM2.5 decreases either, although it may contribute partially. We will later attribute the 2020 decrease at least partially to decreases in activity associated with COVID-19 lockdown.

Fig. 3. Weekly mean raw (purple) and corrected (turquoise) PM2.5 data, as well as weekly mean temperature (°C), wind speed (m s−1), and relative humidity (%) (NOAA data) at the Kinshasa U.S. Embassy site between March 2018 and July 2020.Fig. 3. Weekly mean raw (purple) and corrected (turquoise) PM2.5 data, as well as weekly mean temperature (°C), wind speed (m s−1), and relative humidity (%) (NOAA data) at the Kinshasa U.S. Embassy site between March 2018 and July 2020.

In Fig. 4 we plot the monthly averaged corrected PM2.5 for each of the four other sites (Fig. 1). Data collection began at three of these four sites in November 2019 (UPN, ERAIFT, and Cité Belle Vue), and in February 2020 at the Brazzaville U.S. Embassy, continuing through September 2020. Each of the sites show a similar seasonality, with lower concentrations in November 2019 through April 2020, coinciding with the rainy season. PM2.5 concentrations in June and July are higher at 60–70 µg m–3 from 40 µg m–3 in the rainy season. This is qualitatively consistent with the typical seasonal behavior at the longer-term U.S. Embassy Kinshasa location, though the dry season rebound in PM2.5 in 2020 at the Kinshasa embassy is smaller than at the other four sites. As with the Kinshasa embassy site, corrected PM2.5 levels are nearly always above WHO guidelines, even in the rainy season.

 Fig. 4. Monthly mean corrected data at 4 sites (excluding the Kinshasa embassy) between November 2019 and July 2020. Shaded areas indicate the 95% confidence interval of the monthly average values.Fig. 4. Monthly mean corrected data at 4 sites (excluding the Kinshasa embassy) between November 2019 and July 2020. Shaded areas indicate the 95% confidence interval of the monthly average values.

3.3 Diurnal and Weekly Profiles at Each Sampling Site

We present in Fig. 5 the PM2.5 over the diurnal cycle in each of the Kinshasa-Brazzaville locations using the entire datasets available for each site. Each site tends to have similar diurnal variability, though the magnitude of the PM2.5 concentrations is different at each location. There is typically a very early morning minimum in the diurnal trend at all sites, around 05:00 West Africa Time (WAT), followed by a morning peak at around 08:00. Local traffic emissions and household cooking are expected to be a major contributor to these peaks. Except at UPN-DRC, there is a steady increase throughout the course of the day, as local activity such as cooking and preparing of food, open burning, and other sources begin to contribute to PM2.5. The highest PM2.5 values of the day then peak at about 18:00–20:00, coinciding with evening vehicle traffic and cooking. After 20:00, PM2.5 values start to drop quickly overnight. The highest PM2.5 values occur at the Brazzaville embassy and UPN-DRC, peaking at around 60 µg m–3 in the evening. The lowest PM2.5 values occur at Cité Belle Vue, which is a higher-end residential area where some foreign diplomats tend to live. These areas are less likely to use high-emitting cooking and open burning practices, potentially explaining the lower PM2.5. However, the PM2.5 data at Cité Belle Vue is also only available for the mostly wet season, which can also potentially explain the lower PM2.5 values. The 5-site average is shown in black in Fig. 5 and ranges from about 40 µg m–3 at minimum to above 50 µg m–3 at the peak.

Fig. 5. Diurnal average PM2.5 for the entire data record at each of the 5 sites and the site-wide average (black). Shaded areas indicate the 95% confidence interval of the hourly averages. Hour indicates the local time (West Africa Time).Fig. 5. Diurnal average PM2.5 for the entire data record at each of the 5 sites and the site-wide average (black). Shaded areas indicate the 95% confidence interval of the hourly averages. Hour indicates the local time (West Africa Time).

Average PM2.5 concentration by day of the week is plotted for each site and the area average in Fig. 6. Generally, there are substantial differences in the weekly PM2.5 variation between each location. As in the diurnal variation (Fig. 5), Brazzaville and UPN-DRC have the highest PM2.5 of all the sites, around 50–55 µg m–3. CBV on average is about 10 µg m–3 lower, though this is likely due to a limited data range that mainly includes the wet season. Sunday is generally one of the lowest days for PM2.5 in all sites (except CBV), coinciding with a low point in economic activity during the week. Each site has typically two days of the week in which PM2.5 peaks, though which day varies by site. At UPN-DRC, there is an early week peak Wednesday and then a lower Saturday peak, with a minimum on Thursday. Conversely, the Kinshasa embassy and Brazzaville embassy are highest on Thursday, and ERAIFT-DRC is highest on Friday. This variation among weekday PM2.5 is likely explained by variation in type of location. Residential, educational, and diplomatic locations are represented among the five sites, each of which have different source signatures, emissions patterns, and micrometeorology.

Fig. 6. Day-of-the-week averages for PM2.5 for the entire dataset at each of the 5 sites, and the site-wide average. Shaded areas indicate the 95% confidence interval of the daily averages.Fig. 6. Day-of-the-week averages for PM2.5 for the entire dataset at each of the 5 sites, and the site-wide average. Shaded areas indicate the 95% confidence interval of the daily averages.

Fig. 7 summarizes all daily mean data at all five sites and the 5-site average in a violin plot. Distributions of PM2.5 are mostly unimodal and right-skewed, with the majority of data points located around 50 µg m–3. As previously discussed, UPN-DRC and the Brazzaville embassy have the highest median and quartile ranges for PM2.5. The highest daily mean extreme value (nearly 180 µg m–3) is observed at the Kinshasa embassy. However, the Kinshasa embassy has the most extreme values with several daily means greater than 100 µg m–3, though it also represents the longest data record length. The PM2.5 distributions at every site except for Cité Belle Vue each have “long tails” with nonzero density as concentrations approach 100 µg m–3, indicating a higher frequency of poor air quality at these locations compared to other sites. The 5-site average median daily PM2.5 value is 42.1 µg m–3, a factor of 4 higher than the WHO annual mean guideline of 10 µg m–3.

Fig. 7. Violin plot of daily averaged PM2.5 values for the entire dataset at each location and for the site-wide average. Boxes represent median and interquartile range.Fig. 7. Violin plot of daily averaged PM2.5 values for the entire dataset at each location and for the site-wide average. Boxes represent median and interquartile range.

3.4 Impact of COVID-19 lockdown on PM2.5 in Kinshasa

We use our 29-month dataset (beginning March 2018) at the Kinshasa embassy to assess the potential changes in PM2.5 concentrations attributable to COVID-19-related stay-at-home orders in Kinshasa. The governor of Kinshasa announced a total lockdown for the city starting on 28 March, which was later postponed. Starting 6 April, Gombe, the administrative and commercial center of Kinshasa where the U.S. Embassy is located, was closed for two weeks. We plot and compare the April for 2019 and 2020 in Fig. 8. On average, April 2020 daily mean PM2.5 is about 14.7 µg m–3 lower than the same time period in 2019, about a 40% decrease, larger than the amount of PM2.5 reduction in China during COVID-19 quarantine orders (Shi and Brasseur, 2020). In particular, the evening peak around 20:00 seen clearly in 2019 is mostly absent in 2020. Other than a flattening of the evening peak, the 2020 trend follows 2019 very closely, though offset by at least 10 µg m–3. Natural variability including fire activity may play a role in these decreases, though our preliminary meteorological analysis (see Fig. 3 and associated discussion) suggests that meteorology cannot explain the observed PM2.5 decreases. Daily mean PM2.5 in April 2018 was 3.8 µg m–3 lower compared to April 2019, indicating that there was not an April PM2.5 decreasing trend prior to the COVID-19 pandemic.

Fig. 8. Analysis of PM2.5 changes during COVID-19 at the Kinshasa U.S. Embassy. Averaging time period is 6–20 April. Top: diurnal PM2.5 in 2019 and 2020 by day of the week. Bottom left: diurnal mean PM2.5. Bottom middle: monthly mean PM2.5. Bottom right: day-of-the-week average. Shaded region represents the 95% confidence interval in the mean.Fig. 8. Analysis of PM2.5 changes during COVID-19 at the Kinshasa U.S. Embassy. Averaging time period is 6–20 April. Top: diurnal PM2.5 in 2019 and 2020 by day of the week. Bottom left: diurnal mean PM2.5. Bottom middle: monthly mean PM2.5. Bottom right: day-of-the-week average. Shaded region represents the 95% confidence interval in the mean.

We also analyze satellite observations of AOD at 550 nm wavelength using the MAIAC Level 2 gridded data over land surfaces at 1 km pixel resolution. Fig. 9 shows MAIAC AOD for January–June 2018, 2019, and 2020, and a 2020-versus-2019 difference over the Congo region. AOD levels are elevated over the larger region and especially over Kinshasa and Brazzaville (located around 15.3 E latitude and 4.3 S latitude). Contrary to the surface-level PM2.5 data, AOD was higher in January through July 2019 than 2018. Columnar AOD data is a very different measure than surface-level PM2.5 and do not necessarily vary together (Li et al., 2015; Van Donkelaar et al., 2016). Compared to 2019, AOD in 2020 is about 0.1–0.2 lower than 2019, but only 0.05 lower than 2018. These decreases are qualitatively consistent with our hypothesis of COVID-19 lockdown impacts on air pollution in the Congo region, though could also be consistent with changes in fire activity. Malings, Westervelt, et al. (2020) further explored connections between MAIAC AOD and surface PM2.5 in both Africa and North America, including applications for converting column AOD to surface PM2.5.

Fig. 9. MAIAC Level 2 550 nm AOD at 1 km resolution over the Congo region for a January–July average in (a) 2018, (b) 2019, and (c) 2020, and (d) the difference between 2019 and 2020. Boundaries represent Level 2 administrative boundaries (districts and communes) provided by the United Nations Office for the Coordination of Humanitarian Affairs. The black dashed box indicates the Kinshasa-Brazzaville region and the inset in (d) shows the broader location within the African continent.Fig. 9. MAIAC Level 2 550 nm AOD at 1 km resolution over the Congo region for a January–July average in (a) 2018, (b) 2019, and (c) 2020, and (d) the difference between 2019 and 2020. Boundaries represent Level 2 administrative boundaries (districts and communes) provided by the United Nations Office for the Coordination of Humanitarian Affairs. The black dashed box indicates the Kinshasa-Brazzaville region and the inset in (d) shows the broader location within the African continent.


Every year, air pollution causes millions of premature deaths and a host of illnesses in cities worldwide. Sub-Saharan Africa, where the dearth of air pollution monitoring gives rise to high uncertainty in estimates of exposure and health impacts, is particularly affected. Although the Kinshasa-Brazzaville megalopolis is home to approximately 14 million inhabitants, systematic, long-term air quality data was not publicly available for this region prior to this study. Hence, we deployed five low-cost sensors throughout Kinshasa-Brazzaville in March 2018 and used an MLR model developed for Kampala, a city with comparable environmental conditions, to calculate suitable correction factors. The PM2.5 concentration averaged 46.1 µg m–3 across the monitoring network over the study period, indicating very poor air quality in both Kinshasa and Brazzaville. The levels peaked during the dry season (June–August) and dropped by 5–10 µg m–3 (on average) during the wet season, owing to the dominant role of wet scavenging, but even the latter values were approximately 4 times higher than the WHO guideline.

Furthermore, the decrease in PM2.5 between 2018 and 2020 cannot be ascribed to meteorology, although the paucity of available data limited our analysis. Despite some variation between the study sites, the average concentrations fell within 10 µg m–3 of each other, suggesting a coherent set of sampling locations that represents normal city-wide conditions. Generally, the maximum hourly PM2.5 values occurred around 20:00 WAT, corresponding to the daily early-evening peak in activity. Similarly, the minimum daily values occurred on Sundays, corresponding to the weekly trough in activity. In addition, compared to the same month from the previous year, the daily mean dropped by 14.7 µg m–3 during April 2020—the period of the COVID-19 lockdown—which cannot be explained by meteorological changes alone. Satellite observations of the aerosol optical depth qualitatively confirm our surface-based results.

Our work represents the first step in evaluating the air quality of a fast-growing megacity in sub-Saharan Africa. Future research should address the limitations of this study, including the lack of a local reference-grade monitor in Kinshasa, the absence of robust weather data, and the brevity of the sampling period, as well as explore the potential for better adapting air quality models and satellite observations to Kinshasa and the Congo.


CMM acknowledges a SEAS summer internship and an Earth Institute Student Research Fellowship. DMW acknowledges a Columbia University Climate and Life Fellowship and a Columbia University Climate Center grant. RS acknowledges state assistance managed by the National Research Agency under the Programme d’Investissements d’Avenir under the reference number ANR-18-MPGA-0011 (the Make Our Planet Great Again initiative), as well as equipment support from PurpleAir, LLC. VFM acknowledges the Columbia University Earth Institute for support. CM acknowledges support by an appointment to the NASA Postdoctoral Program at the Goddard Space Flight Center, administered by USRA through a contract with NASA. We acknowledge the following U.S. Department of State diplomats: Jeremy Chen, Dovas Saulys, Lindy Nester, and Thomas Hilleary in Kinshasa and David Kelm in Brazzaville. We acknowledge Paul Green, Engineer Bainomugisha, and the AirQo Kampala project for assistance in retrieving data. The findings of this paper do not represent the views of the U.S. Department of State and are solely the views of the authors. CMM and DMW created figures, led the data analysis and wrote the manuscript. DMW conceived the project and provided equipment. RS provided equipment. PK and RSL provided sensor support. RD provided data quality control and assurance guidance. VFM provided equipment and other guidance. All authors contributed to editing the paper. All low-cost data are publicly available on the PurpleAir website (https://www.purpleair.com/sensorlist). Regulatory instrument data from the U.S. Embassies are available through the U.S. AirNow Department of State data website (https://www.airnow.gov/international/us-embassies-and-consulates/). Satellite data are available from NASA’s Earthdata website (https://earthdata.nasa.gov/).


  1. Amegah, A.K. (2018). Proliferation of low-cost sensors. What prospects for air pollution epidemiologic research in Sub-Saharan Africa? Environ. Pollut. 241, 1132–1137. https://doi.org/10.1016/j.envpol.2018.06.044

  2. Bauer, S.E., Im, U., Mezuman, K., Gao, C.Y. (2019). Desert dust, industrialization, and agricultural fires: Health impacts of outdoor air pollution in Africa. J. Geophys. Res. 124, 4104–4120. https://doi.org/10.1029/2018JD029336

  3. Burnett, R., Chen, H., Szyszkowicz, M., Fann, N., Hubbell, B., Pope, C.A., Apte, J.S., Brauer, M., Cohen, A., Weichenthal, S., Coggins, J., Di, Q., Brunekreef, B., Frostad, J., Lim, S.S., Kan, H., Walker, K.D., Thurston, G.D., Hayes, R.B., Lim, C.C., et al. (2018). Global estimates of mortality associated with longterm exposure to outdoor fine particulate matter. Proc. Natl. Acad. Sci. U.S.A. 115, 9592–9597. https://doi.org/10.1073/pnas.1803222115

  4. Di Antonio, A., Popoola, O., Ouyang, B., Saffell, J., Jones, R. (2018). Developing a relative humidity correction for low-cost sensors measuring ambient particulate matter. Sensors 18, 2790. https://doi.org/10.3390/s18092790

  5. Fiore, A.M., Naik, V., Leibensperger, E.M. (2015). Air quality and climate connections. J. Air Waste Manage. Assoc. 65, 645–685. https://doi.org/10.1080/10962247.2015.1040526

  6. Hagan, D.H., Kroll, J.H. (2020). Assessing the accuracy of low-cost optical particle sensors using a physics-based approach. Atmos. Meas. Tech. 13, 6343–6355. https://doi.org/10.5194/amt-13-6343-2020

  7. He, M., Kuerbanjiang, N., Dhaniyala, S. (2020). Performance characteristics of the low-cost Plantower PMS optical sensor. Aerosol Sci. Technol. 54, 232–241. https://doi.org/10.1080/02786826.2019.1696015

  8. Health Effects Institute (2020). State of Global Air 2019. https://www.stateofglobalair.org/report (accessed 26 August 2020).

  9. Holstius, D.M., Pillarisetti, A., Smith, K.R., Seto, E. (2014). Field calibrations of a low-cost aerosol sensor at a regulatory monitoring site in California. Atmos. Meas. Tech. 7, 1121–1131. https://doi.org/10.5194/amt-7-1121-2014

  10. Jacob, D.J., Winner, D.A. (2009). Effect of climate change on air quality. Atmos. Environ. 43, 51–63. https://doi.org/10.1016/j.atmosenv.2008.09.051

  11. Jayaratne, R., Liu, X., Thai, P., Dunbabin, M., Morawska, L. (2018). The influence of humidity on the performance of a low-cost air particle mass sensor and the effect of atmospheric fog. Atmos. Meas. Tech. 11, 4883–4890. https://doi.org/10.5194/amt-11-4883-2018

  12. Jiang, Y., Zhou, L., Raghavendra, A. (2020). Observed changes in fire patterns and possible drivers over Central Africa. Environ. Res. Lett. 15, 0940b8. https://doi.org/10.1088/1748-9326/ab9db2

  13. Jiao, W., Hagler, G., Williams, R., Sharpe, R., Brown, R., Garver, D., Judge, R., Caudill, M., Rickard, J., Davis, M., Weinstock, L., Zimmer-Dauphinee, S., Buckley, K. (2016). Community Air Sensor Network (CAIRSENSE) project: Evaluation of low-cost sensor performance in a suburban environment in the southeastern United States. Atmos. Meas. Tech. 9, 5281–5292. https://doi.org/10.5194/amt-9-5281-2016

  14. Johnson, K.K., Bergin, M.H., Russell, A.G., Hagler, G.S.W. (2016). Using low cost sensors to measure ambient particulate matter concentrations and on-road emissions factors. Atmos. Meas. Tech. Discuss. https://doi.org/10.5194/amt-2015-331

  15. Kelleher, S., Quinn, C., Miller-Lionberg, D., Volckens, J. (2018). A low-cost particulate matter (PM2.5) monitor for wildland fire smoke. Atmos. Meas. Tech. 11, 1087–1097. https://doi.org/10.5194/amt-11-1087-2018

  16. Kelly, K.E., Whitaker, J., Petty, A., Widmer, C., Dybwad, A., Sleeth, D., Martin, R., Butterfield, A. (2017). Ambient and laboratory evaluation of a low-cost particulate matter sensor. Environ. Pollut. 221, 491–500. https://doi.org/10.1016/j.envpol.2016.12.039

  17. Levy Zamora, M., Xiong, F., Gentner, D., Kerkez, B., Kohrman-Glaser, J., Koehler, K. (2019). Field and laboratory evaluations of the low-cost plantower particulate matter sensor. Environ. Sci. Technol. 53, 838–849. https://doi.org/10.1021/acs.est.8b05174

  18. Li, J., Carlson, B.E., Lacis, A.A. (2015). How well do satellite AOD observations represent the spatial and temporal variability of PM2.5 concentration for the United States? Atmos. Environ. 102, 260–273. https://doi.org/10.1016/j.atmosenv.2014.12.010

  19. Lobo, K.K., Tshiashala, M.D., Lumu, B.M., Matamba, K., Ronneau, C. (1990). Air quality status in kinshasa as determined by instrumental neutron activation analysis, atomic absorption spectrometry, and ion-exchange chromatography, in: Nuclear Analytical Methods in the Life Sciences, Humana Press, pp. 295–305. https://doi.org/10.1007/978-1-4612-0473-2_33

  20. Lyapustin, A., Wang, Y., Korkin, S., Huang, D. (2018). MODIS collection 6 MAIAC algorithm. Atmos. Meas. Tech. 11, 5741–5765. https://doi.org/10.5194/amt-11-5741-2018

  21. Magi, B.I., Cupini, C., Francis, J., Green, M., Hauser, C. (2020). Evaluation of PM2.5 measured in an urban setting using a low-cost optical particle counter and a Federal Equivalent Method Beta Attenuation Monitor. Aerosol Sci. Technol. 54, 147–159. https://doi.org/10.1080/02786826.2019.1619915

  22. Malings, C., Tanzer, R., Hauryliuk, A., Kumar, S.P.N., Zimmerman, N., Kara, L.B., Presto, A.A. (2019). Development of a general calibration model and long-term performance evaluation of low-cost sensors for air pollutant gas monitoring. Atmos. Meas. Tech. 12, 903–920. https://doi.org/10.5194/amt-12-903-2019

  23. Malings, C., Tanzer, R., Hauryliuk, A., Saha, P.K., Robinson, A.L., Presto, A.A., Subramanian, R. (2020a). Fine particle mass monitoring with low-cost sensors: Corrections and long-term performance evaluation. Aerosol Sci. Technol. 54, 160–174. https://doi.org/10.1080/02786826.2019.1623863

  24. Malings, C., Westervelt, D.M., Hauryliuk, A., Presto, A.A., Grieshop, A., Bittner, A., Beekmann, M. (2020b). Application of low-cost fine particulate mass monitors to convert satellite aerosol optical depth to surface concentrations in North America and Africa. Atmos. Meas. Tech. 13, 3873–3892. https://doi.org/10.5194/amt-13-3873-2020

  25. Mbelambela, E.P., Hirota, R., Eitoku, M., Muchanga, S.M.J., Kiyosawa, H., Yasumitsu-Lovell, K., Lawanga, O.L., Suganuma, N. (2017). Occupation exposed to road-traffic emissions and respiratory health among Congolese transit workers, particularly bus conductors, in Kinshasa: A cross-sectional study. Environ. Health Prev. Med. 22, 11. https://doi.org/10.1186/s12199-017-0608-9

  26. Subramanian, R., Kagabo, A.S., Baharane, V., Guhirwa, S., Sindayigaya, C., Malings, C., Williams, N.J., Kalisa, E., Li, H., Adams, P. (2020). Air pollution in Kigali, Rwanda: spatial and temporal variability, source contributions, and the impact of car-free Sundays. Clean Air Journal 30, 1–15. https://doi.org/10.17159/caj/2020/30/2.8023

  27. Shi, X., Brasseur, G.P. (2020). The response in air quality to the reduction of Chinese economic activities during the COVID‐19 outbreak. Geophys. Res. Lett. 47, e2020GL088070. https://doi.org/10.1029/2020GL088070

  28. Tryner, J., L’Orange, C., Mehaffy, J., Miller-Lionberg, D., Hofstetter, J.C., Wilson, A., Volckens, J. (2020). Laboratory evaluation of low-cost PurpleAir PM monitors and in-field correction using co-located portable filter samplers. Atmos. Environ. 220, 117067. https://doi.org/10.1016/j.atmosenv.2019.117067

  29. Tryner, J., Quinn, C., Windom, B.C., Volckens, J. (2019). Design and evaluation of a portable PM2.5 monitor featuring a low-cost sensor in line with an active filter sampler. Environ. Sci. Process. Impacts 21, 1403–1415. https://doi.org/10.1039/c9em00234k

  30. Van Donkelaar, A., Martin, R. V., Brauer, M., Hsu, N.C., Kahn, R.A., Levy, R.C., Lyapustin, A., Sayer, A.M., Winker, D.M. (2016). Global estimates of fine particulate matter using a combined geophysical-statistical method with information from satellites, models, and monitors. Environ. Sci. Technol. 50, 3762–3772. https://doi.org/10.1021/acs.est.5b05833

  31. Westervelt, D.M., Horowitz, L.W., Naik, V., Tai, A.P.K., Fiore, A.M., Mauzerall, D.L. (2016). Quantifying PM2.5-meteorology sensitivities in a global climate model. Atmos. Environ. 142, 43–56. https://doi.org/10.1016/j.atmosenv.2016.07.040

Share this article with your colleagues 


Subscribe to our Newsletter 

Aerosol and Air Quality Research has published over 2,000 peer-reviewed articles. Enter your email address to receive latest updates and research articles to your inbox every second week.

79st percentile
Powered by
   SCImago Journal & Country Rank

2023 Impact Factor: 2.5
5-Year Impact Factor: 2.8

Aerosol and Air Quality Research partners with Publons

CLOCKSS system has permission to ingest, preserve, and serve this Archival Unit
CLOCKSS system has permission to ingest, preserve, and serve this Archival Unit

Aerosol and Air Quality Research (AAQR) is an independently-run non-profit journal that promotes submissions of high-quality research and strives to be one of the leading aerosol and air quality open-access journals in the world. We use cookies on this website to personalize content to improve your user experience and analyze our traffic. By using this site you agree to its use of cookies.