Enhancing the precision of low-cost pm2.5 air pollution sensors through the application of machine learning | |
| Author | Asamaporn Punkru |
| Call Number | AIT Thesis no.EV-24-01 |
| Subject(s) | Air--Pollution--Thailand--Data processing Machine learning Air--Pollution--Measurement |
| Note | A thesis submitted in partial fulfillment of the requirements for the degree of Master of Science in Environmental Engineering and Management |
| Publisher | Asian Institute of Technology |
| Abstract | Air pollution poses a significant challenge in Thailand, particularly with the annual recurrence of PM2.5, which comprises tiny particles that are detrimental to health. PM2.5, defined as particulate matter with an aerodynamic diameter of 2.5 µm or less, is a pervasive issue affecting air quality. Installing air quality monitoring stations offers a viable solution to address this concern. Nevertheless, the substantial expenses and operational complexities associated with these stations have left certain regions in Thailand without access to critical monitoring infrastructure. Consequently, Low-Cost Sensors (LCS) have emerged as an alternative for air quality monitoring due to their compact size, portability, and affordability. However, LCS performance can be significantly affected by high variability in conditions of elevated relative humidity. This study aims to improve the correctness of concentration of PM2.5 from LCS by using machine learning methods i.e. Linear regression, Muti-Linear Regression, Decision Tree Regression, Random Forest Regression, and XGBoost Regression models. This study analyzed secondary data from LCS deployed co-located with the reference air quality monitoring stations at Bangkok, Chiang Mai, and Ubon Ratchathani provinces, Thailand. The performance metric was used in this study following United States Environmental Protection Agency (USEPA) standards such as R-squared (R 2 ) and Root Mean Square Error (RMSE). The R 2 of LCS namely Aerosure, Airenvi, Doophoon and Dustboy before adjusting data are 0.88, 0.87, 0.55 and 0.89, and RMSE are 13.39, 22.92, 32.73 and 20.77 µg/m3 respectively. After the sensors are calibrated by using models, the PM2.5 concentration from the LCS is closer to concentration from the reference instruments, especially with the XGBoost model adjustment. This model provides better concentration adjustment than other models because XGBoost has a R2 closer to 1 (Aerosure = 0.9, Airenvi = 0.89, Doophoon = 0.79 and Dustboy = 0.9) that means LCS reading PM2.5 typically more precise, and they are provides lower RMSE than unadjusted data (Aerosure = 12.22 µg/m3 , Airenvi = 19.16 µg/m3 , Doophoon = 21.9 µg/m3 and Dustboy = 16.39 µg/m3 ). A lower RMSE can be defined as less error between the sensor measurements and the reference instrument measurements. In this study, it was also found that meteorological data improves sensor accuracy because the concentration from LCS with meteorological data adjustment has better performance metrices than the concentration without meteorology data. The percent relative bias was used in this part to determine the sensors' measurement error when compared to the reference instrument, and they can define over or underestimate values of PM2.5. The result shows that weather conditions affect sensor readings, especially Relative Humidity (RH), which causes high variability in sensor readings. When RH is over 81%, PM2.5 of sensor readings increase significantly, and relative bias increases as well. Additionally, sensors tend to underestimate values in the range of 0–25 µg/m3 and provide concentration close to the reference instrument in the range of 25.1–75 µg/m3 . The PM2.5 concentration reading sensor is overestimates above 75 µg/m3 , resulting in high variability and a high relative bias. Moreover, this study provided a brief guideline following The Enhanced Air Sensor Guidebook from USEPA to develop guidelines for adjusting PM2.5 readings from the LCS in Thailand. This would be highly beneficial for environmental agencies to develop better policies and interventions for air quality management. |
| Year | 2024 |
| Type | Thesis |
| School | School of Environment, Resources, and Development |
| Department | Department of Energy and Climate Change (Former title: Department of Energy, Environment, and Climate Change (DEECC)) |
| Academic Program/FoS | Environmental Engineering and Management (EV) |
| Chairperson(s) | Ekbordin Winijkul;Thongchai Kanabkaew (Co-Chairperson) |
| Examination Committee(s) | Cruz, Simon Guerrero;Adisorn Lertsinsrubtavee |
| Scholarship Donor(s) | Her Majesty the Queen's Scholarships (Thailand) |
| Degree | Thesis (M. Sc.) - Asian Institute of Technology, 2024 |