As of mid-day yesterday (7/31/24) we have rolled out a new “correction factor” for our air quality readings. Why do we need to correct the readings from our PurpleAir PA-II sensor? Because it tends to over count particulate matter (PM2.5) by a significant margin (~30-40%).
Previously we were using the EPA conversion (Barkjohn 2021) that was designed for wildfire smoke. It’s definitely an improvement over the default factory algorithm (CF_ATM) for outdoor measurements. However, the EPA conversion still uses the proprietary “CF1” or “CF_ATM” formulas created by the sensor manufacturer, Plantower. Apparently, these calculations do not correspond to known physical processes or to any known standard for calculating PM2.5.
That is why the lead author of several foundational studies involving thousands of PurpleAir sensors and the chief proponent of the ALT-CF3 correction, retired environmental research scientist Lance Wallace (and colleagues), devised another method. Rather than rely on “black box” numbers, the ALT-CF3.4 leverages the various raw particle “bins” (PM0.3, PM0.5, PM0.5, PM1.0, PM2.5) created by the twin laser counters.
The bottom line is that this algorithm, as applied to our particular sensor (Plantower 5003), should get us to within several percentage points of the gold standard federal reference method (FRM) for measuring air quality. As a bonus, at particularly low concentrations which are common in Upper Michigan for much of the year, this method should produce more granular readings rather than just going to zero (reaching the so-called “limit of detection”) as happened too often over the past year since launching AQI on the website.
Also, as of today, we have rolled out an additional data field in the weather history tables. If you click on the “Air Quality” tab, you can now view not only uncorrected AQI and adjusted AQI but also corrected PM2.5 readings. These calculations relate back to the correction factor discussed above. Researchers and (nerdy?) enthusiasts usually deal in particulate concentrations rather than AQI.
I should also mention — and, yes, thank you for reading this far — that the measurements appearing on the home page and 24-hour history graph are rolling 10-minute averages. We do this to suppress transient readings (smoking passersby, smoking BBQs and fire pits, sensor noise, etc). This provides the right balance between instantaneous responsiveness and rock-you-to-sleep readings (hourly averages).
Okay, I’ll say it. This is yet another exclusive feature offered by our website. Do not try this on your “weather app”.
