Comparison of Physics-Based and Machine Learning Based Approach for Calibration of Low-cost Particulate Matter Sensors

Brijal Prajapati, MANORANJAN SAHU, Chandra Venkataraman, Pratim Biswas, Indian Institute of Technology Bombay

     Abstract Number: 697
     Working Group: Instrumentation and Methods

Abstract
Low cost particulate matter sensors are receiving significant attention as they can be used in large number for spatial and temporal measurement of PM mass and number concentration. However, the data reliability is questionable as these sensors are affected by numerous parameters such as temperature, relative humidity, and hygroscopicity of particles. To ensure accurate and reliable measurement of particulate matter concentrations, performance evaluation and calibration of LCS co-located with reference instrument at site is essential. In this study, the performance of the low-cost sensor (APT Maxima) was evaluated with the reference instrument SASS (Speciation Air Sampling System) sampler for developing suitable calibration factor that include impact of the meteorological parameter such as relative humidity and temperature, and hygroscopicity. In this study, we have demonstrated a systematic physics-based method for calibration of LCS based on κ-Köhler theory and Mie theory. For comparison with statistical models, linear regression and machine learning algorithm were also applied. In physics-based model, calculated total light scattered intensity shows good linearity with reference PM2.5 measurements. Physics based model performed better for both the sites as compared to MLR, kNN, RF, and GB ML algorithms with R2, RMSE, and MAE values of 0.72,18.21, and 13.36 for Bhopal site, and 0.91, 7.84, and 5.76 for Kashmir site, respectively. Study indicates that physics-based approach for LCS calibration is suitable and can be transferable to different sites.

Keywords: Low-cost sensor, Physics-based model, Sensor calibration

Acknowledgements: This work was supported by the Ministry of Environment, Forest and Climate Change (MoEF&CC), Government of India, under the NCAP-COALESCE project {Grant No.14/10/2014-CC (Vol.II)}.

References
[1] Badura, M., Batog, P., Drzeniecka-Osiadacz, A., & Modzel, P. (2019). Regression methods in the calibration of low-cost sensors for ambient particulate matter measurements. SN Applied Sciences, 1(6).
[2] Dharaiya, V. R., Malyan, V., Kumar, V., Sahu, M., Venkatraman, C., Biswas, P., Yadav, K., Haswani, D., Raman, R. S., Bhat, R., Najar, T. A., Jehangir, A., Patil, R. D., Pandithurai, G., Duhan, S. S., & Laura, J. S. (2023). Evaluating the Performance of Low-cost PM Sensors over Multiple COALESCE Network Sites. Aerosol and Air Quality Research, 23(5), 220390.
[3] Kumar, V., & Sahu, M. (2021). Evaluation of nine machine learning regression algorithms for calibration of low-cost PM2.5 sensor. Journal of Aerosol Science, 157, 105809.