In order to check the accuracy of AirVisual devices, AirVisual conducted research between the 1st of June to the 30th of June 2015. In the comparison study, the AV monitor was located near the Beta Attenuation Monitor (US Embassy Beijing), meaning that the monitors were not sampling exactly the same air. During the comparison, the correlation (r squared) between government BAM monitor and Air Visual monitor was 0.959, for hourly data the correlation (r squared) was 0.83.
Other accuracy measurements against reference sensors have shown similar high correlation levels.
monitors that are set up to provide public data are validated before
publishing. The monitor installations are visually verified through photo
submissions, to confirm that the sensor is in a suitable location (e.g. not in
close proximity to hyperlocal emission sources).
The data is validated for a period of 7-10 days before publishing, by checking different parameters, for example that PM2.5 levels do not fluctuate irregularly, and are comparable to nearby/regional data where available.
AirVisual Machine Learning algorithms then takes care of validating and calibrating the sensors based on temperature, humidity and the local pollution composition.