Calibration of PurpleAir monitors

For the last few years, I’ve been analyzing the performance of the PurpleAir monitors. Last week the Sensors journal published my detailed comparison of the Plantower CF_1 algorithm with an alternative algorithm called ALT-CF3.

This is Open Access. There is also a link within the pdf that takes you to the Supplemental Information (pictures of the PurpleAir and SidePak research-grade monitors, for example.)

ALT-CF3 is a based on a method that has been used for decades to calculate PM2.5 from monitors providing estimates of particle numbers in several size categories. 33 PurpleAir monitors within 500 m of 27 Federal Reference monitors in California supplied the data over 18 months resulting in the choice of 3.0 as the calibration factor (CF) (Wallace et al., 2021).

Some good news from the new paper in Sensors is that the ALT-CF3 algorithm produces extremely good precision, averaging 4-8%, compared to about 10-15% for the CRF_1 algorithm.

More good news is that the Limit of Detection (LOD) of the ALT-CF3 algorithm is below 1 ug/m3, compared to the CF_1 LOD of 3-5 ug/m3. Since outdoor PM2.5 is often below 3-5 ug/m3, this means that a rather high percentage (sometimes more than half) of outdoor measurements using the CF_1 algorithm are below the LOD.

Even more good news is that the monitors showed sustained high performance for the entire 3 years of the study.
A great advantage of the PurpleAir monitors is that they can be used indoors to provide the first actually measured long-term exposures to PM2.5 that we have. However, indoor concentrations are usually lower than outdoors, meaning that an even higher percentage of CF_1 measurements will fall below the LOD.

A drawback to using the CF_1 or CF_ATM algorithms is that concentrations falling below a certain value are all given a value of zero. Statisticians do not agree with this approach, preferring that the reported concentrations not be altered. At low concentrations, zeros often predominate. The ALT-CF3 algorithm NEVER reports zeros, since there are always some particles in the 0.3-0.5 um size category.
In the Map page, there are five alternative conversions offered, such as ALT-CF3, LRAPA, EPA, etc. However, four of these depend on the CF_1 algorithm, and thus have the same problems with the poor precision, prevalence of zeros at low concentrations, etc. due to the CF_1 algorithm. The only conversion that does not depend on CF_1 is the ALT-CF3 version.

Fortunately, the ALT-CF3 version is available for downloading on the PurpleAir API website. There it is named PM2.5 alt.
Lance Wallace
Lwallace73@gmail.com

9 Likes

The link to the paper on calibration of PurpleAir monitors referred to above is

1 Like

The calibration discussed above is only for the PA-II monitors with PMS 5003 sensors. Also, it is only for outdoor air in California. This leaves us with some unfinished business:

  1. Do we need a separate calibration for outdoor air in other areas of the country?
  2. Do we need a separate calibration for wildfire smoke?
  3. Do we need a calibration for indoor air?
  4. Do we need a calibration for the PA-I monitor with the PMS 1003 sensor.

My colleagues at Stanford and I actually made a start on 3) above-- the calibration for indoor air. We published a study of the source strength of vaping marijuana in 2020

We found that a single puff of marijuana vapor had about 2-6 times as much PM2.5 as a single puff of a tobacco cigarette. We also found, by comparing the PurpleAir monitors used in the study to co-located SidePak research monitors, (and then to gravimetric pump-filter-microbalance methods) that the best calibration factor for the marijuana vapor (mixed with outdoor air penetrating the exposure chamber (a room in my house)) was again 3.0. So we have one estimate of the calibration factor for a single indoor source. Once we go through another thousand or so sources we might have a useful indoor air calibration factor. This study required 124 6-hour experiments over the course of one year. We have our work cut out for us and our descendants!

REgarding other unfinished business, for problem 1) above the EPA studied about 6 geographci areas in the US and came up with a single “nationwide” calibration,with a correction for RH. However, their correction factor depends ultimately on the CF_1 system and therefore has the same faults as the CF_1 system.

A recent study by Liang found a new calibration factor for wildfires, but again it depended on the CF_1 method.

https://www.pnas.org/doi/10.1073/pnas.2106478118

However, in this case, the concentrations were so high that we need not worry about zeros appearing very often, so their calibration factor (0.52) is probably pretty good. It suggests that the normal overestimate of 50% or so by the CF_1 method increases to nearly a factor of 2 overestimate for wildfire smoke.

3 Likes

Here is an easy way to calculate the ALT cf=3 PM2.5 value from the particle numbers given in the first four size categories in the “Secondary” part of the download using the widget in the Map page and selecting the “No” conversion. Recall that these categories are the numbers (per deciliter but that won’t matter) of all the particles greater than the given category. Therefore the first category 0.3-0.5 um is all particles greater than 0.3 um To get the number of particles in that category alone, you must subtract the number of particles >0.5 um (given in the next larger category of 0.5-1 um) from the number greater than 0.3 um. This is the number of particles in the smallest size category 0.3-0.5 um. By repeating this for the next two size categories you get the number of particles in each of the first three categories. There are also some steps putting in the density and calculating the masses of the particles in each size category, but all that is contained in the three coefficients shown below. Suppose N1 is the number of particles/dl in the smallest size category (0.3-0.5 um), and N2 and N3 are the number in the next two categories up to 1-2.5 um. Now the total mass will be given by the following equation:

PM2.5 = 3 (0.00030418*N1 + 0.0018512 *N2 + 0.02069706 * N3)

The factor of 3 is the calibration factor.

For those persons using the API site, you should get exactly the same values using the PM2.5 alt variable.

1 Like

Hi Lance,

I’m working on a new manuscript and was hoping to reference your explanation of the Alt-CF3.4 correction model. I have two questions:

1. What is the best reference to cite for this correction model?
2. Is the PM_Alt available for download via the PurpleAir API the same as the Alt-CF3.4 model you mentioned here?

Thanks,
Masoud

1 Like

Hi Masoud–

Good to hear from you again.

  1. I will check my refs for the best one explaining the ALT CF3.4

  2. The API variable is called pm2.5 alt. (or perhaps pm2.5 alt3.4) and it is exactly the same as the alt CF3.4 algorithm. PurpleAir recently updated it so that uses a calibration factor of 3.4 instead of the older value of 3.0.

2 Likes

Hi Lance,

Thanks for the prompt response. I downloaded the data for PM_Alt about 6 months ago and I think it’s for the calibration factor of 3.0. The manuscript is almost at the final stage, so I may not be able to re-download the updated data (for the calibration factor of 3.4). I think the formula for the old PM_Alt (version 3.0) is
PM2.5 = 3 (0.00030418*N1 + 0.0018512 *N2 + 0.02069706 * N3), but I was hesitant about which reference is the best for this correction model.

BTW, I appreciate it if you could send me one reference for each of the calibration factors (3.0 and 3.4).

Thanks in advance,
Masoud

2 Likes
  1. Best reference is probably
    Wallace, L. (2024). Measured M2.5 indoors and outdoors related to smoking prevalence by Zip code using 14,400 low-cost monitors in California, Washington,and Oregon. Indoor Environments 1 (2024) 100043.

Should be Open Access.

1 Like

Thanks, Lance, for the information.

Do not use the DustTrak. It reports PM2.5 2.5x higher than a filter reference method. Do not use the TAPI-640 either, same problem with woodsmoke. The BAM measures PM2.5 properly regardless of composition.

1 Like

Multiple studies have used EPA gravimetric or equivalent methods to calibrate PurpleAir monitors. Some of these are mentioned in the Map page, including the EPA formula and the pm2.5 alt algorithm

1 Like

I want to make a quick note on this. The pm2.5_alt field still returns Alt-CF3 data for backward compatibility. However, you can retrieve Alt-CF3.4 data by using a pipe and specifying the constant. In this case, it would be: pm2.5_alt|3.4.

We have plans to add this to the documentation, but for now, you can find an official note about it here: What is the Difference Between CF=1, ATM, and ALT?

2 Likes

Dear Eyasu Lake

The CF_1 and CF_ATM algorithms were supplied by the manufacturer of the sensors used in the PurpleAir monitors.l Many authors have found that these algorithms overestimate PM2.5 by about a factor of 2.. References to those papers can be found in the first 11 or so papers in the attached link..

The AQI algorithms are exceedingly misleading. The AQI is properly defined as a daily (midnight to midnight) average of PM2.5 that is then assigned an AQI value correlated with the AQI breakpoints, such as the 12 (or 9) ug/m3 annual standard. It is NOT defined as related to an instantaneous reading by any monitor. PurpleAir folks have been made aware of this point, but I imagine they have not yet decided how to deal with the problem.

Therefore all four of your algorithms are not suited for use.

The best algorithm for indoor air is the pm2.5 alt algorithm, available at the PurpleAir API site.

For outdoor air, my recommendation continues to be the pm25 alt algorithm, which does not include any correction for T or RH. T has never been found to have much effect. RH has a known effect on certain molecules at high values (>60-70%). In that case, the EPA algorithm attempts to correct for RH. However, the advantage of using the same algorithm for indoor and outdoor air (to determine such things as the infiltration factor) may outweigh the possible improvement concerning corrections at high RH outdoors. (Indoors,the RH almost never reaches 70%.) The two largest studies do not correct for RH (Wallace, Indoor Air; 2022; (forgot the name, begins with E, PLOS 2023 or 2024).

For outdoor ,I recommend collecting both the PM2.5 alt and EPA algorithm. Compare them and if there is amajor difference, you can go with the EPA algorithm, but if it is minor, use the ALT algorithm for both indoor and outdoor values.

(Attachment bibliography with links to last 11 papers 8-30-2023.docx is missing)

2 Likes

Hi @Lance
I am doing a project on Air Quality in my local area & am recovering the raw PM2.5_ATM data from my PurpleAir Flex’s SD Card.
Are you able to please tell me what the formula is for the ALT_CF3.4 correction so I can correct the raw data?
I have tried looking at the research paper that is on the PurpleAir map, but have had no success finding the formula.

Thank you so much :slightly_smiling_face:

1 Like

Hi @Lance I am GINO from Lima Peru. Here we have an extremly humidity in summer (between 60 and 70% percent) In winter +90%. Wich algorithm have to choose… ATL 3.4 or US EPA? I can tell when the humidity is close to +80% the pm2.5 goes up quickly. We are next to the pacific ocean and the wind blows all the time from the sea. If i turn off the conversion; in this moment, pm2.5 marks 22. If i turn on in ALT3.4 pm2.5 marks 14 and in US EPA 12. The HR is 85% since 3 hours ago. DO i have to choose NO conversion, US EPA or ALT3.4 for a better medition from my station?
Thank you for your post my purple air station is @ClimaPeruMundo in Lima Peru

1 Like

Gino–

“No conversion” means you are actually using the original algorithm called “CF_ATM” supplied by the manufacturer of the two sensors in PurpleAir monitors. This algorithm is known to overestimate PM2.5 by about 50% (multiple studies such as AQ-SPEC, Barkjohn etc.). So indeed if it reads 22 ug/m3, then both ALT3.4 and EPA will read that much lower (14 or 12 ug/m3). So all three of your readings basically agree, when you allow for the overestimate of the CF_ATM (“no conversion” algorithm. In other words, you should probably never use the “no conversion” algorithm.

Regarding humidity, the ALT3.4 algorithm does not include a correction for RH. The EPA algorithm multiplies the “no conversion” CF_ATM value by 0.5 and then adjusts for relative humidity. However, this algorithm is based on US data alone and may not be that useful in the extreme conditions in Peru. So in general you may wish to use the EPA algorithm at high levels of RH. However, there is some uncertainty about the actual RH value to use. That is, the RH as determined by the PurpleAir monitors is reduced by about 15% at low RH and up to 25% at high RH. So do you use the RH as determined by the PurpleAir monitor using the Bosch (?) temperature-RH instrument in the PurpleAir monitor or by some different co-located accurate RH measurement. This gets even more complicated by not knowing which RH measurement to use, because different particle compositions may lead to the particle immediately reacting to the interior RH or not; in the latter case it “carries with it” the very much lower RH observed in the ambient air.

1 Like

AirFright:

This may be a question to put to the PurpleAir support staff (e.g. Andrew White). My guess is that the older Flex monitors may have been using the original calibration factor of 3.0 for their PM2.5 ALT algorithm. In that case, the proper way to correct it to the ALT 3.4 value would simply be to multiply the PM2.5 ALT value by 3.4/3.0, about a 12 or 13% correction. More recently, the API allows you to download the ALT3.4 correction by adding a suffix "|3.4"to your request. But again, I am not familiar enough with the SD Card to tell you the exact way to obtain the correction.

Regarding the formula for obtaining the ALT3.4 value, I do have such a formula, but it depends on the number of particles in each of the three smallest categories and would thus complicate your downloads. I believe the PurpleAir staff has applied that formula correctly for both the 3.0 and 3.4 calibration factors, saving you the additional effort (and $!)

2 Likes

[quote=“Lance, post:19, topic:482, full:true”]
AirFright:

This may be a question to put to the PurpleAir support staff (e.g. Andrew White). My guess is that the older Flex monitors may have been using the original calibration factor of 3.0 for their PM2.5 ALT algorithm. In that case, the proper way to correct it to the ALT 3.4 value would simply be to multiply the PM2.5 ALT value by 3.4/3.0, about a 12 or 13% correction. More recently, the API allows you to download the ALT3.4 correction by adding a suffix "|3.4"to your request. But again, I am not familiar enough with the SD Card to tell you the exact way to obtain the correction.

Regarding the formula for obtaining the ALT3.4 value, I do have such a formula, but it depends on the number of particles in each of the three smallest categories and would thus complicate your downloads. I believe the PurpleAir staff has applied that formula correctly for both the 3.0 and 3.4 calibration factors, saving you the additional effort (and $!)

2 Likes

See below for my answer

2 Likes

Thank you very much for the response, Lance. I am an amateur meteorologist and have semi-professional stations. I also have a Purple Air device and a Xiaomi Mi AirQuality. I’ve had the Xiaomi Mi for over 10 years, and its measurement is almost exact with the US EPA conversion. I think the extreme humidity in Lima (located in a desert near the Pacific Ocean) increases PM2.5 values. For example, today in a coastal little town near Lima, the PM2.5 measurement is 30; a very high value since there’s almost no human movement in that area. I believe it’s the extreme RH (a constant +85%). The entire Peruvian coast is an extreme desert and extreme HR.
Would you agree to use the US EPA conversion for my Purple Air station?

1 Like