Preprints
https://doi.org/10.5194/amt-2021-282
https://doi.org/10.5194/amt-2021-282

  29 Oct 2021

29 Oct 2021

Review status: this preprint is currently under review for the journal AMT.

Machine Learning Techniques to Improve the Field Performance of Low-Cost Air Quality Sensors

Tony Bush1,2, Nick Papaioannou1, Felix Leach1, Francis D. Pope3, Ajit Singh3, G. Neil Thomas4, Brian Stacey5, and Suzanne Bartington4 Tony Bush et al.
  • 1Department of Engineering Science, University of Oxford, Parks Road, Oxford, OX1 3PJ, UK
  • 2Apertum Consulting, Harwell, Oxfordshire, UK
  • 3School of Geography, Earth and Environmental Sciences, University of Birmingham, Edgbaston, Birmingham, B15 2TT, UK
  • 4Institute of Applied Health Research, University of Birmingham, Edgbaston, Birmingham, B15 2TT, UK
  • 5Ricardo Energy and Environment, The Gemini Building, Fermi Avenue, Harwell, Didcot, OX11 0QR, UK

Abstract. Low-cost air quality sensors offer significant potential for enhancing urban air quality networks by providing higher spatio-temporal resolution data needed, for example, for evaluation of air quality interventions. However, these sensors present methodological and deployment challenges which have historically limited operational ability. These include variability in performance characteristics and sensitivity to environmental conditions. In this work, field 'baselining' and interference correction using Random Forest regression methods for low-cost sensing of NO2, PM10, and PM2.5 is investigated. Model performance is explored over 7 months obtained by field deployment alongside reference method instrumentation. Workflows and processes developed are shown to be effective in normalising variable sensor baseline offsets and reducing uncertainty in sensor response arising from environmental interferences. A mean absolute error of 2.5 ppb, 4.8 µg/m3 and 2.9 µg/m3 for NO2, PM10, and PM2.5 respectively, was achieved for corrected field-deployed sensors compared to a reference method. When used to correct data collected under environmental conditions outside model training, results meet European data quality objectives, albeit with lower accuracy than data from within the trained range.

Tony Bush et al.

Status: open (extended)

Comment types: AC – author | RC – referee | CC – community | EC – editor | CEC – chief editor | : Report abuse
  • RC1: 'Comment on amt-2021-282', Anonymous Referee #1, 19 Nov 2021 reply
    • AC1: 'Reply on RC1', Tony Bush, 24 Nov 2021 reply

Tony Bush et al.

Tony Bush et al.

Viewed

Total article views: 532 (including HTML, PDF, and XML)
HTML PDF XML Total BibTeX EndNote
427 100 5 532 3 3
  • HTML: 427
  • PDF: 100
  • XML: 5
  • Total: 532
  • BibTeX: 3
  • EndNote: 3
Views and downloads (calculated since 29 Oct 2021)
Cumulative views and downloads (calculated since 29 Oct 2021)

Viewed (geographical distribution)

Total article views: 507 (including HTML, PDF, and XML) Thereof 507 with geography defined and 0 with unknown origin.
Country # Views %
  • 1
1
 
 
 
 
Latest update: 08 Dec 2021
Download
Short summary
Poor air quality is a human health risk for which there is demand for evidence to inform policies to improve it. Low-cost sensors are attractive for this with advantages over traditional methods but they also bring challenges – sometimes data quality can be low. In this study in Oxford, we develop and test Machine Learning methods to improve NO2, PM10 and PM2.5 sensor data quality. We show that with these methods low-cost sensor data quality can be greatly improved, in line with EU objectives.