Recent research using Praxis/Urban air quality instruments has investigated the use of ML data correction methods to improve field performance of these devices, which use low-cost air quality sensors. The research, published just this month, covers sensing data for NO2, PM10 and PM2.5, read the paper here.
The challenge for using low-cost sensors
At South Coast Science, we have long understood the potential of instruments designed with low-cost sensors for local authority and corporate projects alike. Because they are more affordable and portable than reference instruments, they can be used to create high density networks. This allows data to be gathered across a network of devices within a city in real-time, generating a detailed picture of local air quality patterns. These can then be mapped to changing traffic conditions, for example.
However, uncertainty in data accuracy has stalled the widespread adoption of these instruments. UK manufacturer Alphasense is the ubiquitous supplier of low-cost sensors across almost all of the industry to develop ambient air quality monitoring devices. There’s a reason for this – they have excellent repeatability and are easy to use. Unfortunately the sensors are highly sensitive to variations in temperature and relative humidity, so they alone can’t provide high levels of data confidence when used outdoors.
Using real-world data to train data-correction models
The OxAria group and authors of this study have demonstrated that ML data correction techniques can be used to improve the accuracy of low-cost sensors. The paper demonstrates that the ML methods can reduce the error in NO2, PM10 and PM2.5 data by up to 88% – 95%.
Above – figure 2: Graph showing instrument readings over 7 months. Black line shows reference method readings, the blue shows uncorrected sensor readings, green shows corrected sensor readings. The green and black lines are almost on top of each other.
The team used data gathered from sensors operating in the field over a seven month period, collocated with reference instruments at the Defra monitoring station at St Ebbe’s in Oxford. Some data is used to train the data correction model (the ML part) and some data is used to validate the corrected sensor data.
This echoes the methodology for field-testing and validation work by South Coast Science, where large air quality data sets have been gathered by deploying Praxis units around the globe. The real-world data is used to train the ML algorithm. The corrected data is then validated by collocation with reference instruments with partners like Ricardo and the UN.
Above: 5 graphs showing different stages of sensor correction model, with comparison to reference method readings. As each sensor correction is applied the output matches more closely with the reference instrument output.
* Reference: Bush, T., Papaioannou, N., Leach, F., Pope, F. D., Singh, A., Thomas, G. N., Stacey, B., and Bartington, S.: Machine learning techniques to improve the field performance of low-cost air quality sensors, Atmos. Meas. Tech., 15, 3261–3278, https://doi.org/10.5194/amt-15-3261-2022, 2022.
This website uses cookies so that we can provide you with the best user experience possible. Cookie information is stored in your browser and performs functions such as recognising you when you return to our website and helping our team to understand which sections of the website you find most interesting and useful.
Strictly Necessary Cookies
Strictly Necessary Cookie should be enabled at all times so that we can save your preferences for cookie settings.
If you disable this cookie, we will not be able to save your preferences. This means that every time you visit this website you will need to enable or disable cookies again.