Hi Adrian and all PurpleAir staff
I now know how Plantower calculates PM1 and PM2.5. I also know why they have zeros.
My paper (attached) supporting these claims was accepted today by Science of the Total Environment. In the next few days I expect I can provide a link to the published paper.
I claim that they calculate the number of particles in the smallest 3 size categories just as I do in the ALT-CF3 algorithm. However, instead of calculating PM2.5 by multiplying each size category by a different constant value determined by fitting to some calibration aerosol (which in their case they do not identify), they first add the particle numbers in the two smallest size categories (0.3-0.5 um and 0.5-1 um). They do this for both their PM1 and PM2.5 estimates in their CF1 algorithm… In addition, I believe they add a small constant value on the order of -1 ug/m3 to get their algorithm to go to zero when the true concentration is zero.
Using this model of their approach and testing it by results from four PurpleAir monitors collecting both indoor and outdoor data for 6 months, I can get nearly perfect estimates of their CF1 results.
Figure 5. The fit to the CF_1 algorithm for PM2.5 for sensor 1a using a fixed general model for all sensors: PM1 = 0.0042*(N1+N2) + 0.93*N3 -1.17.
Using this model which includes a value of -1.17 ug/m3 to be subtracted from every measurement, there were 18,000 negative results out of about 100,000 two-minute average measurements. All 18,000 cases were identified as zero in the CF1 results.
Comparing to the collocated ALT-CF3 estimates (using the recent estimate of 3.4 for the calibration factor), the CF1 algorithm overestimated PM2.5 by about 32%.
By the way, I would be delighted to be proved wrong, because the only way to do that would be for Plantower to tell us what their real algorithm is.
(Attachment Cracking the Code–final manuscript.docx is missing)