Characterising low-cost sensor performance for the design and development of a low-cost, high-precision, multi-parameter clustered sensor instrument for outdoor air quality monitoring.

Low-cost sensors (LCS) for the detection of atmospheric composition are being increasingly used for monitoring air quality. Increasing the number of measurements locations in an air quality network can be useful for the validation of atmospheric models and provide improved estimates of personal pollution exposure. Performance of LCS, relative to existing reference instruments, has been seen to be highly variable, but there are currently no formalised standards or certified calibration procedures for their use. Within this project, laboratory studies and field-testing were undertaken to characterise the performance of several commercially available LCS. The sensors tested prioritised those atmospheric pollutants that are regulated under the UK and EU legislation, e.g. nitrogen dioxide. A range of sensor technologies, including electrochemical and metal oxide sensors has been evaluated. Clustering multiple identical sensors was an effective approach that improved data quality and reduced the required frequency of calibration with co-located reference instruments, also improving medium frequency noise and sensor reproducibility. New approaches to resolving sensor chemical cross-interferences were explored, from simple linear regression to machine learning algorithms. This improved the agreement between sensors and reference instruments in the laboratory and field. Clusters of sensors were built into a multi-pollutant instrument which was deployed in various locations to investigate sensor performance in different environments. Through the application of machine learning over all the sensor signals, it was possible to produce a signal that was close to the reference measurements, indicating that LCS can be used in a similar manner to an air quality monitoring station. One implication of this is that LCS can be used over the short term (weeks - months) to complement the existing networks by increasing the number of ground observations, which would facilitate the interpolation of pollutant concentration gradients between relatively sparse network stations to better estimate pollution maps.