PurpleAir vs EPA sensors

It seems that there are frequent questions about why a particular PurpleAir sensor or group of sensors don’t seem to agree with an EPA sensor. Besides the technical differences in the sensors, in these discussion I’m interested that I’ve never seen (and certainly may have missed…) a discussion about local environmental effects.

I live in the mountains where such effects can be quite exaggerated and consequently easier to notice. Within a short walking distance there can be temperature variations of 5 or 10 degrees with similar changes in humidity (and using my lungs as sensors, particulate pollution as well). These topoclimatic effects can be seen on even smaller scales around here and I imagine even within a very small location at a particular dwelling (open air vs sheltered in the back under an eave or whatever)

There aren’t many PurpleAir sensors in our area, but I frequently notice a very significant difference between our sensor and those few which are in the valley, which is about 6 miles down the mountain and from the local EPA sensors.

So, it doesn’t seem to me in general, recognizing the technical differences, that there should be much surprise that any given PurpleAir sensor might be different, sometimes significantly from the local EPA reports.

I would love to hear from or be pointed to someone who has a real understanding of these effects (in contrast to my own which are uneducated observations).

Thanks

2 Likes

A very astute observation. The best way to look for local effects would be to locate the nearest EPA sensor (either FRM or FEM monitor), 10-20 of the nearest outdoor PurpleAir monitors, and collect data for (best case) a year or so to allow for seasonal effects.

This was actually carried out over a 5-year period (2017-2021) for 10,000 outdoor and 4400 indoor PurpleAir monitors in the West Coast states… However, the purpose was not to focus on local effects but to calibrate the PurpleAir monitors to the mean EPA sensors. This resulted in the pm2.5 alt (aka ALT CF-3) algorithm now available for downloading. However, the extensive data could be analyzed to look at variations around this mean, but that has not been carried out.

Thank you.

The density of monitors in our area is pretty low so tracking local monitors here wouldn’t be very helpful. On a very quick count there are about 15 senors in the immediate 1200 sq miles which includes one EPA sensor.

In the study you reference, there’s certainly an impressive amount of data. It would have been interesting if the data/evaluation had included included those local effects. During the data collection and evaluation process in that study was some of the other data available and evaluated in these monitors such as humidity, operating temperature, etc.?

I think one could have an interesting conversation about the attempts at correction of the sensor data. Though not a very solid data point, I find that clinically, how i feel, i.e. how the air quality affects my breathing, correlates nicely with the raw PM2.5 and less well with the ALT CF-3 correction.

something you can consider is making sure you’re looking at the correct averaging intervals. for example, most air agencies (& EPA AirNow maps) report out 1-hour averaged data, and there is a slight delay in updating each hour of new data on the map. Also, AirNow is displaying something called the NowCast, which is a rolling calculation that takes into account previous hours and trend analysis. It’s typically not a straight value reported from a concentration.

If you’re looking at real time or even 10-minute data from PurpleAir sensors, you’re bound to get very different results than EPA monitors that are reporting NowCast hourly values. Also, PurpleAir sensors report higher concentrations than regulatory monitors, but applying the USEPA correction factor can help address the over-estimation bias and get you closer readings.

I am not sure if you’re referring to ‘EPA Sensors’ as the regulatory equipment, or PurpleAir sensors owned & operated by EPA.

Thanks - I think what I was originally getting at in my original post and what I am more focused on is the idea that variability in correlation between sensors shouldn’t be surprising. It seems to be related to some degree, large or small, to local effects and that this idea isn’t commonly discussed. In my own very limited experience and from an even smaller (maybe microscopic) knowledge base, these topoclimatec effects can be pretty large, depending upon geography, vegetation or even density and type of building construction. A great example for me of local variability this time of year is in watching a high wind gust blow snow across our field where the actual heterogeneity of air movement can be easily visualized in the swirling snow patterns in that concentrated space with areas of high density and with areas of apparent voids.

We recently had a fire 2 miles from my house and the purple air AQI was elevated the whole time and went up to 2,500, but the airnow.gov in the same vicinity said it was green/under 50 the whole time

Hi Steph, I’m guessing that you were in the smoke plume from that metal recycler. I hope you were able to evacuate! I also hope you are taking advice on how best to deal with the smoke particles that been deposited on your house’s curtains and walls, and on the ground outside. There are no easy answers to this; but I think it important to do what you can, and try to comfort yourself with the fact that you and your loved ones are probably still much more likely to die in an auto accident, or even by gunfire, than from a cancer which was caused by any of the bazillions of possibly-carcinogenic chemicals which may have been released by this fire… or by their cumulative assault on your body.

I live in Auckland NZ but have a brother in Pasadena CA who lives about 5 miles from Altadena. I had suggested (in my geeky way, which he appreciates – but I realise it’s a rare taste!) that he order some swab-test strips for lead and cadmium. These are cumulative toxins which might have been present in the Altadena smoke… and which also might have been present in the smoke from the Camden recycler filre. But! It’s not easy to interpret the results of a swab test – even harder than trying to “correct” the readouts of a PurpleAir monitor so that they’d (somehow! impossibly!) be an accurate guide to an US EPA PM2.5 reading. So I think you’d be best advised to take advice from your municipal authorities. My heart goes out to you!

Yes indeed, smoke plumes can be quite localised, and if there’s not much wind they’ll dissipate rather slowly. Accordingly, the location of the particulate sensor is a very important factor. The EPA sensors are much more expensive to purchase than a PurpleAir sensors and the EPA sensors are carefully calibrated (which requires expert labour – another expense), The EPA sensors are mounted by experts in appropriate positions – so that their readouts aren’t invalidated by local effects – and they are maintained. So there can’t be very many of them.

As you seem to be aware, the PurpleAir sensors work on quite a different principle than the EPA sensors. It’s not possible for the two types of sensors to give comparable readings unless the colour of the particulates is known. Smoke from wildfires that burn one sort of vegetation in some particular way is rather different to smoke from another wildfire, or from the tailpipe of a diesel engine. It’s impossible to adjust the readout from a PurpleAir sensor to line up accurately with the readout from an EPA particulate sensor – except on a single, well-characterised, type of airborne particulate.

However the inaccuracies aren’t so large as to invalidate the important contribution that the prolific PurpleAir sensors provide, to people (including researchers) who are interested to know “where” a smoke plume is currently located, roughly “how fast” it is dissipating, “where” it is moving, and (if it was a point source) “where” it started.

There are nowhere near enough EPA sensors to give this fine-grained information.

And the fact that a PurpleAir Sensor’s readout is quite biased on certain types of smoke, and the fact that some PurpleAir Sensors produce biased readouts because they weren’t installed according to the manufacturer’s guidelines, doesn’t prevent us from using their data. Yeah it’s “dirty” but that’s the reality in the world that all “big data” researchers inhabit – just gotta do your best to avoid getting misled by the most prevalent forms of “dirt”!

The bias in a PurpleAir sensor’s readout (in comparison to an EPA sensor’s readout) is quite high in the case of windblown dust from the dry bed of the Owens Lake, according to one research study. By a factor of 5 or 6. Wow. But even in this case: you could trace the contours of the dust-plume from the dry bed of what was once Owens Lake much much more accurately by deploying 100 of these highly-biased PurpleAir sensors than you could hope to do using data from the very sparse EPA sensor network.

I see little hope of improving on the correction method of Barkjohn et al. (2021) or of the correction method adopted in 2021 for the U.S. EPA’s AirNow network. They’ll both give quite biased results on some types of particulates. They’re “one size fits all” methods, without any consideration of the particulate’s colour distribution, size distribution, and its rate of clogging the sensor’s innards. Furthermore, they’ll both have validity issues whenever the RH is high. If two sensors use lasers of different wavelengths (optical, or infrared) their raw-count differentials will be quite dependent on the colour of the particulates.

All to say that I think it’s a hiding to nowhere to attempt to “correct” a PurpleAir readout, except in a context where you’re pretty confident the elevated PM2.5 is mostly due to woodsmoke, or to diesel smoke, or to dust from (say) a nearby dry lakebed. In such contexts I think you could hope for a valid cross-calibration. And why bother? The PurpleAir readouts are quite indicative of what an EPA device would measure, and will be very helpful for fine-grained tracking of smoke plumes in areas where there are multiple PurpleAir sensors. There are only a handful of these in Auckland NZ, where I live, but even so I’m pretty confident I can tell the difference between an hour in which my sensor is mostly picking up woodsmoke from a neighbour’s smoky fireplace, and an hour in which my sensor’s readout is lagging the readouts from folks who live closer to the harbour. Ocean going freighters sometimes spew quite a bit of diesel smoke, it would seem, and when the wind blows toward me from the harbour my PurpleAir Sensor is sensitive enough to detect those harbour-sourced particulates… and I really can’t hope to make an accurate estimation of how much PM2.5 I’m breathing, in a year, from woodsmoke and diesel smoke. It’s enough to be noticeable but I’m almost-surely gonna die from something other than emphysema or asthma or lung cancer which (arguably!) was caused by this smoke.

Thank you so much for your kind response. We haven’t been given any guidance by municipal authorities. I wil lookup ideas for how to deal with the deposited particles and welcome any suggestions.