Air quality sensors… can you simply plug and play?

Air quality sensors… can you simply plug and play?
09 March 2022

Air quality sensors… can you simply plug and play?

The ability to take air pollution measurements has never been easier. There are many manufacturers and resellers of sensor systems, ranging from low-cost personal measurement devices like the Plume Labs Flow, through to more expensive, larger multi-pollutant monitoring stations like the Bosch Immission Monitoring Box.

Perhaps more important than the ability to take measurements is being confident in the value of the resulting data. While sensors are easy to acquire, deriving value from the datasets in relation to real world ambient conditions remains a tricky task.

This raises a number of questions:  

  • How can I be sure that my system works?
  • Can I just install it and leave it to collect data?
  • Can I use sensors instead of reference monitoring stations/diffusion tubes?
  • Can I use them for mobile studies/schools/hotspot identification?
  • How confident can I be that the measurements made by a sensor system are actually of any use?

At this point in time there’s no magic bullet. Advice is emerging from expert groups, such as the USEPA, WMO and AQEG but most sensor systems still do not have formal performance certification for ambient air quality measurements. And when suppliers are charging many hundreds or thousands of pounds for a system, you want to have some confidence that what you’re about to buy actually works, right?

[Not a typical monitoring station!]
image of reference monitoring station


For a number of years, with the cooperation of a selection of manufacturers, businesses and local authorities, we have installed sensor systems at “reference” monitoring stations (similar to the image above) for evaluation. This approach allows us to see how the equipment performs in the real world and to give manufacturers an opportunity to refine and improve the performance of their sensor systems. Ultimately, the end goals are continuous improvement for the manufacturers and, for our clients, knowledge of what systems can be used with confidence.

To date, we’ve investigated the performance of sensor systems from suppliers including:  

  • Aeroqual
  • Airly
  • Alphasense
  • AQMesh
  • Clarity
  • EarthSense
  • PlumeLabs
  • South Coast Science
  • Vaisala
  • Vortex

In this and subsequent blogs, we’ll try to answer some of the questions relating to sensor use to give you a better idea of the potential pitfalls, and the right questions to ask suppliers.

1.    Do sensors actually work?
The best way to answer this question is to run the system in a location for which you already have some air quality measurement data. Visually comparing a sensor timeseries with a quality dataset is a good way to confirm it’s working.

[Click diagram to view full size image]


The plot above shows how an NO2 sensor performs next to a reference analyser. You can see a strong correlation between the two datasets – they track similarly. But does that mean all monitoring needs are solved?

Well, kind of, but not really… This is only the first part of the story.

The chart clearly shows there is an obvious step (offset) between the two datasets – indicating a margin of error associated with the sensor reading. There may also be a need for a calibration factor to be applied, i.e. a value which the sensor reading has to be multiplied with to get the corresponding true reference or the correct reading.

To find this, we need to put the data in a scatter plot, where each data pair is plotted as a single point.

 
[Click diagram to view full size image]


This plot gives us lots of useful information. Firstly, it confirms and quantifies the offset (+9.6) – resolving the uncertainty associated with the sensor reading. Next, it tells us that there’s a calibration factor that needs to be applied (the sensor data needs to be divided by 0.809 after the offset is subtracted). This results in the two datasets closely matching each other (the value of r2 is high at 0.87) and more importantly, the corrected figures provide data with smaller errors than the raw datasets.

Only after the raw data is adjusted, can we be confident that this particular sensor system “works”.

2.    What happens if I don’t have a reference station to put my sensor system next to?
Things get a bit tricky here. You can still compare sensor measurements with a neighbouring reference station, but it’s not ideal. The reference and sensor locations must be the same type of location, e.g. both roadside or both residential. The trends should be similar, but you should expect some differences in time-series data, and scatter plots will be of less value. It will be difficult to calculate a calibration factor and offset for your sensor system that gives you reasonable data quality, but there are possible approaches that can be used to address this problem (more about this later).

3.    Can I just install it and leave it to collect data?
The short answer is no. The longer answer is it depends on what system you buy. Ricardo has undertaken evaluation studies on different sensor systems over the years, and there is often quite a bit of variation in performance.

 
[Click diagram to view full size image]


The above plot shows how NO2 measurements from a series of notionally identical sensor systems can vary out of the box. By and large, there is reasonable agreement between most of the sensors, but three or four of the systems plotted here are periodically performing differently compared to the others. It would be a significant risk to trust that the sensor system will just work out of the box, and you should never just “fit and forget”.

Of course, you can use the knowledge gained from asking questions 1 and 2 to look at relationships between sensor systems and reference analysers to improve confidence in your measurements.

 
[Click diagram to view full size image]


This plot shows the results from three scaled sensors located at different locations close to a school in Oxford. Roadside, school gates and playground locations were chosen to see if any differences in measurements could be seen (red = roadside, orange = gates, green = playground) when compared to a reference background location (in black) in the city.

You can see that the trends between the sensors and the reference station match well, and the roadside is higher than the other two sensor locations.

Taken in context, all three systems have worked well for this study.

4.    Can I use sensors to replace reference monitoring stations and diffusion tubes?
This depends on a number of factors: why you are making measurements, whether your data is critical for a statutory air quality review and assessment, and/or whether you are simply undertaking research to get a better understanding of spatial distribution etc.

It is unlikely that the data quality from a sensor system will be good enough for it to replace reference measurements, at least in the medium term. But, with appropriate quality control, sensor systems could well find a use alongside diffusion tubes, providing useful data about short-term variations in pollution concentrations.

5.    Can I use sensors for mobile/school studies/hotspot identification?
Yes, if you take care to ensure that your sensor(s) provide meaningful data as described above. The smaller systems are usually battery or USB powered, often with GPS and data connectivity (or can connect to a mobile phone for these services), so it is possible to move around with them to assess personal exposure or to identify the hotspots in an area. Similarly, the ease of deployment and immediate data availability make sensor systems ideal to build into KS1/2/3 teaching programmes.  We’ve collaborated on these sorts of programmes at a number of schools in Scotland, for example.

The above gives you a taste for the sorts of challenges you can expect when you take the plunge and buy a sensor system. Hopefully, it is clear that, at least for now, there is a lot of work required to turn your raw data into something usable. In the next blog, we’ll explore this in a bit more detail, looking at some recent comparison data and hint at what might be on the horizon for QC of sensor data.
 

Could your organisation benefit from sensor data correction?


Ricardo routinely provide sensor manufacturers and users with technical evaluation, guidance and development support in terms of dataset management, quality assurance and quality control (QA/QC) and processing. For more information, visit our QA/QC sensor data correction services page.

For direct queries or to discuss how Ricardo could help your organisation improve the quality of its sensor data reporting, please get in touch using the form on this page.