Low-Cost Sensor Performance Intercomparison, Correction Factor Development, and 2+ Years of Ambient PM2.5 Monitoring
Garima Raheja1,2, James Nimo3,4, Emmanuel K-E Appoh5
1Department of Earth and Environmental Sciences, Columbia University, New York, New York 10027, United States.
Environmental Science & Technology
|July 12, 2023
Summary
Low-cost sensors show strong correlation with reference monitors for particulate matter (PM2.5) in Accra, Ghana. Data correction models improve accuracy, revealing concerning PM2.5 levels that necessitate air quality mitigation strategies.
Area of Science:
- Environmental Science
- Air Quality Monitoring
- Public Health
Background:
- Particulate matter air pollution is a major global mortality cause, with significant data gaps in low- and middle-income countries (LMICs).
- Low-cost sensors offer a potential solution for ambient air monitoring in under-monitored regions, but their performance and intercomparison in Africa are understudied.
Purpose of the Study:
- To conduct the first intercomparison of different brands of low-cost particulate matter (PM2.5) sensors in Africa.
- To evaluate statistical and machine learning models for correcting low-cost sensor data.
- To assess ambient PM2.5 concentrations in Accra, Ghana, using a network of low-cost sensors.
Main Methods:
- Colocation of QuantAQ Modulair-PM, PurpleAir PA-II SD, and Clarity Node-S Generation II monitors with a reference-grade Teledyne monitor in Accra, Ghana.
- Intercomparison of sensor performance and evaluation of four models (Multiple Linear Regression, Random Forest, Gaussian Mixture Regression, XGBoost) for data correction.
- Deployment and data correction using Gaussian Mixture Regression for a network of 17 Clarity Node-S monitors from 2018 to 2021.
Main Results:
- All tested low-cost PM2.5 sensors strongly correlated with the reference monitor but showed a high bias for Accra's ambient pollution.
- QuantAQ Modulair-PM exhibited the lowest mean absolute error (3.04 μg/m³), followed by PurpleAir PA-II (4.54 μg/m³) and Clarity Node-S (13.68 μg/m³).
- XGBoost model showed the best performance for data correction (R²: 0.97, 0.94, 0.96), but Gaussian Mixture Regression was preferred for out-of-range data.
- The corrected network data revealed a daily average PM2.5 concentration of 23.4 μg/m³ in Accra, 1.6 times the WHO guideline.
Conclusions:
- Low-cost sensors are viable for monitoring PM2.5 in LMICs like Ghana, but require careful calibration and data correction.
- Accra's average PM2.5 levels exceed WHO guidelines, indicating a need for prompt air quality mitigation strategies.
- Continued monitoring and development of air quality management plans are crucial for rapidly growing African cities.
