I appreciate this manuscript because it’s important that the air quality research community and users of low-cost air quality monitors, like the PurpleAir, understand that these devices can underestimate airborne dust concentrations by a lot.

My biggest concern is that the authors do not use the most recent U.S. EPA PurpleAir correction equation. Equation 1 (on Line 91) is the U.S.-wide correction equation that was published by Barkjohn et al. in Atmospheric Measurement Techniques (DOI: 10.5194/amt-14-4617-2021) but, since the time of that publication, the EPA has updated their PurpleAir correction approach because (a) Equation 1 underestimated very high smoke concentrations (as the authors note here) and (b) limitations associated with the PurpleAir API made a correction equation based on the PMS5003-reported PM_2.5 ATM concentrations more appealing. The updated EPA correction is a piecewise equation that (a) uses the PM_2.5 ATM value as a predictor (instead of the PM_2.5 CF=1 value) and (b) predicts high concentrations of wildfire smoke more accurately. See additional information in this presentation: https://cfpub.epa.gov/si/si_public_record_report.cfm?dirEntryId=353088&Lab=CEMM. EPA indicates that they started using the correction shown on Slide 14 in Summer 2021.  The authors should (a) confirm with EPA that they are still using the piecewise correction shown in Slide 14 in this presentation, (b) confirm with EPA whether the piecewise correction shown on Slide 14 is the “EPA” correction being used on the PurpleAir map, and then (c) re-do their analysis using whichever correction EPA is currently using on the PurpleAir map and the AirNow Fire and Smoke map. 

Additional comments:

2. Methods, Part I: What geographic area and date range did the authors’ analysis span? Did the authors look for all pollution events at all U.S. regulatory monitoring sites that fell within a specific date range and had PM_2.5 concentrations > 40 µg m^-3 for 3 consecutive hours? Were the regulatory sites with PM_2.5 > 40 µg m^-3 identified first and then the nearest PurpleAir monitors identified? Did the authors impose any limit on how close the PurpleAir monitor had to be to the regulatory monitor for data from the two to be compared? What range of correlation coefficients between regulatory and PurpleAir data would have been accepted?  What software did the authors use to analyze their data? Were pollution events identified systematically (this would be preferred!) or manually?
3. Methods, Part II: What type of regulatory monitor is being used at the Keeler, CA site? The EPA map indicates that it’s a TEOM. Please note the regulatory monitor make and model in the text. How far from the regulatory monitor was the PurpleAir monitor?
4. Lines 224-225: “These differences are statistically significant for urban versus dust, but not for smoke versus dust at the 95% confidence level.” Are the authors sure that the “size distribution” data reported by sensor aren’t influenced by the absolute PM concentrations reported by the sensor? What do the PAS 0.3 µm/5 µm count ratios look like when high concentrations of smoke are present? Could the approach proposed to correct data from dust events lead to some smoke events being mischaracterized as dust events?

Technical corrections:

Lines 73-74: I think there are some words missing from this sentence. I suggest revising as something like “The PM_2.5/PM₁₀ ratio for smoke is also similar to that for urban pollution: 0.55–0.75 (Xu et al., 2017).”