Nationwide Assessment of Ambient Particulate Matter in Madagascar Using Low-Cost Sensors and Statistical Modeling
Authors: Oladimeji Mudele¹,²*, Eleanor Klibaner-Schiff³*, Giacomo De Nicola¹,², Marissa L. Childs⁴, Hervet J. Randriamady¹,⁶, M. Ando Miharifetra⁵,⁶, Joël Rajaobelison⁷, Martin Bbaale⁸, Gideon Lubisia⁸, Joel Ssematimba⁸, Deo Okure⁸, Engineer Bainomugisha⁸, Christopher Golden¹,²,⁶
¹ Department of Nutrition, Harvard T.H. Chan School of Public Health, Boston, MA, USA
² Department of Environmental Health, Harvard T.H. Chan School of Public Health, Boston, MA, USA
³ Department of Human Evolutionary Biology, Harvard University, Cambridge, MA, USA
⁴ Department of Environmental and Occupational Health Sciences, University of Washington, Seattle, WA, USA
⁵ Departement Production et Partenariat, Institut Malgache des Vaccins Veterinaires (IMVAVET), Ampandrianomby, Antananarivo, Madagascar
⁶ Madagascar Health and Environmental Research (MAHERY), Maroantsetra, Madagascar
⁷ Institut National des Sciences et Techniques Nucl´eaires, Madagascar
⁸ AirQo, Makerere University, Kampala, Uganda
- These authors contributed equally as first authors.
Ambient air pollution is a major cause of premature mortality in Madagascar, yet routine ground-based monitoring has been absent. We deployed 20 low-cost AirQo sensors across five administrative regions in diverse ecological zones after calibrating the sensors using collocation with reference-grade monitors in Boston and U.S. Embassy in Antananarivo, Madagascar. We found strong temporal agreement (R² values of 0.88 and above) between AirQo sensors and co-located reference monitors that supports their use for population-level exposure assessment. Using hourly data aggregated to daily means, we characterized spatial and temporal patterns of fine (PM2.5) and coarse (PM10) particulate matter, evaluated exceedances of World Health Organization (WHO) air quality guidelines, and fit generalized additive mixed models (GAMMs) to quantify meteorological drivers of daily variability.
Annual mean PM2.5 concentrations at all sites exceeded the WHO annual guideline, and WHO 24-hour PM2.5 exceedances were frequent, especially in the South and in Antananarivo. Diurnal PM patterns showed morning and evening peaks consistent with traffic and cooking patterns, and PM concentrations were highest during the dry season when biomass burning, low wind speeds, and limited rainfall coincide. Higher wind speeds and more humid conditions were associated with lower particulate levels, while higher daily maximum temperatures were positively associated with particulate matter concentrations.
These results demonstrate that validated low-cost sensor networks can deliver policy-relevant air quality evidence in data-scarce settings and reveal that PM2.5 in Madagascar systematically exceeds WHO guidelines. The findings identify biomass burning, residential biomass combustion, and traffic emissions as key intervention targets and provide a foundation for integrating air quality into Madagascar’s emerging climate-smart public health initiatives.
This repository contains data, scripts, and figures for the analysis of ambient particulate matter in Madagascar using low-cost sensors and statistical modeling.
- ClimateVariables.R: Loads and cleans merged sensor data, merges with climate variables (precipitation, wind, temperature), and prepares data for modeling.
- collocationPM2.5.R: Performs collocation analysis between AirQo sensors and reference monitors in Boston and Anatananarivo.
- EDA.R: Conducts exploratory data analysis, including diurnal, weekly, and seasonal patterns of PM2.5 and PM10, and visualizes trends by device and region.
- GAMM.R: Fits generalized additive mixed models (GAMMs) to quantify meteorological drivers of daily PM variability.
- GAMM_pm25_pm10_pvalues.R: Loads data, cleans data and fits GAMM model.
- KMeans.R: Performs clustering analysis on daily PM2.5 averages to identify spatial patterns and sensor groupings.
- Manual_EDA.R: Processes and analyzes manually collected data from three sites (Marofototra, Antaravato, Vinanibe), including hourly and daily summaries and heatmaps.
- data_explo.R: Merges all API data files, joins with device metadata, and prepares the main merged dataset for analysis.
- plot_map.R: Visualizes sensor locations and regional boundaries on a map of Madagascar using spatial data.
- forest_plot.R: Generates forest plots of model estimates and confidence intervals for key predictors.
For more details on data sources and usage, see the comments in each script and the data directory structure.