The PurpleAir (and similar low-cost PM) sensors have received a lot of attention recently, due to their popularity and wide deployments worldwide. Many publications have shown the utility of these sensors for air quality applications, including (at the risk of tooting my own horn) several of my own papers (e.g. Rose Eilenberg et al. https://doi.org/10.1038/s41370-020-0255-x). A number of laboratory and field evaluations have shown the limitations of these sensors, especially the fact that the size distribution reported by the Plantower sensor is unrealistic (e.g. Kuula et al., Tryner et al., Zou et al.) and that the Plantower sensor is a nephelometer, not a particle counter despite manufacturer claims (He, Kuerbanjiang, & Dhaniyala). Bernd Laquai (2017) had shown the lack of response of a similar low-cost PM sensor to coarse aerosols https://www.researchgate.net/publication/320707986_Particle_mass_distribution_dependent_inaccuracy_of-low_cost_sensors_Bjhike_HK-A5 ; I have seen similar results in our testing in Doha, Qatar (hopefully, to be published soon!)

Hagan and Kroll have also published an optical model for various low-cost OPCs https://amt.copernicus.org/articles/13/6343/2020/amt-13-6343-2020.html 

However, this paper makes a unique contribution in that they measure the insides of the sensor, develop a nice model specific to the PurpleAir PMS5003 that predicts the nephelometric response compared to a perfect nephelometer, and use that to show that the Plantower effectively cannot detect supermicron particles due to the optical configuration, with a smaller contribution from the flow path (which was my default assumption). These results likely also explain the laboratory test results across a range of PM sensors (Plantower, Nova, Sensirion, etc.) by Kuula et al. (though the Omron remains a mystery - it uses a thermal heater for convective flow with a relatively straight path, and was responsive to coarse mode aerosol). The long-term evaluation at Mauna Loa and Boulder further enhance the value of this paper. Hence, I recommend publication after minor revisions.

Specific comments:

1. Please use PM_2.5 throughout, not the eyesore that is "PM2.5".

2. Line 178: Is the 9.4 cm³ volume for one half of the sensor or both halves?

3. Lines 201-204: This seems a noteworthy finding - that even at low wind speeds, particles larger than 2 µm may not even make it inside the sensor - with losses of ~90% for 5 µm particles. These losses, combined with the inefficient scattering due to the optical configuration, likely explain Kuula's results. On the other hand, the Omron sensor in Kuula's testing detected coarse particles, and that flow path is more straightforward (uses convective heating for aspiration).

4. Line 223 - it would appear (given the aspiration efficiency above) that this statement "most particles larger than 10 µm are lost" could be made for particles larger than 2 µm? 

5. Sec 3.3 - it is not clear how (or if) the model accounts for the non-ideal response of the photodiode - SI suggests it does not. Please explain how this may affect your results.

6. Fig 7 - this figure might be better shown (or additionally shown) without the normalization to 0.1 µm particle. It will be interesting to know the absolute effect of polarization and and PA geometry on scattering at a given diameter without this normalization.

7. Sec 4.3 - PurpleAir monitors come in a plastic housing and are not heated (except for some internal heating by the electronics board). Since Mauna Loa and Boulder are both low-RH environments usually, the effect of the heater modification introduced by the authors is unclear. Can you provide a comparison of the RH in the heated and unheated PurpleAir with ambient RH? Does heating affect the plastic housing of the PurpleAir e.g. offgassing/melting?

8. Can you compare the NRMSE values with uncertanties reported in previous PurpleAir evaluations? e.g. in Malings et al. (2020), we report a mean absolute error of ~4 µg/m³ for hourly average PM_2.5 (average PM_2.5 in Pittsburgh is ~10 µg/m³). Others have published similar values as well.

9. It would be good to place the results in Sec 5.2.5 PA-PMS size distribution in the context of previous results by Kuula et al., Tryner/Volckens, Zou/May, and He/Dhaniyala.

https://amt.copernicus.org/articles/13/2413/2020/

https://doi.org/10.1016/j.jaerosci.2020.105654

https://doi.org/10.1080/02786826.2021.1905148

https://doi.org/10.1080/02786826.2019.1696015

10. Lines 859-860: It has been shown (e.g. Malings et al.) that the as-reported PM_2.5 mass concentrations (effectively CH1) are ~double the reference PM_2.5 values. However, the authors here speculate that the unmodified PurpleAir may underestimate high-RH, low visibility aerosol scattering coefficients. (Is there an unheated nephelometer, given the heat of the halogen lamp?) If I understand correctly, the authors are comparing PA response to scattering coefficient at ambient high RH. It might help to make this more explicit, e.g. state that "previous studies have reported overestimated PM_2.5 mass concentrations by the PurpleAir at high RH, but that is when the PA is compared to a reference monitor reporting dry aerosol concentrations - which is quite different than our application of the PA as an aerosol light scattering instrument".

The paper “Evaluating the PurpleAir monitor as an aerosol light scattering instrument” by Ouimette et al, examines the possibility of using Purple Air PMS sensor data to determine integrated aerosol light scattering coefficient.  A model considering Mie theory and the sensor geometry is used to predict light scattering signals expected from the sensor and the forward and backward scattering truncation.  The model is used predict sensor performance as a function of particle size and the results confirm that the sensor does not measure size distributions.  And that the signal is proportional to scattering coefficient.

The paper presents a comprehensive picture of the working of PMS5003.  The sensor details, model results and experimental validation adds to the existing knowledge on PMS 5003 and critically confirms findings of other studies that have concluded that the sensor behaves more like a nephelometer rather than a scattering spectrometer.  The paper is well written and its findings are likely to be very useful to the growing community of scientists using these sensors for air quality measurements.

I have minor points for the authors to consider.

Lines 106-108 – “is of light scattered by particles (Kelly et al., 2017) which traditionally has been … using integrating nephelometers”.  This sentence should be reworded.  As it reads currently, it seems like light scattering measurements are only made by nephelometers.  

Lines 137-139.  “Model predictions are then compared with yearlong field data at NOAA’s Mauna Loa Observatory …”.  Please clarify exactly what predictions are compared with what data.

Lines 140-141: “… an empirical relationship is developed to estimate the light scattering and uncertainty from the PA-PMS data.”  Light scattering intensity?  And uncertainty of what?

How is the uncertainty in the physical geometry and optical geometry accounted for in the model?

In Figure 2, the precision is shown as a function of concentration.  How much of the decrease in precision with decreasing concentrations can be explained by Poisson statistics of number of particles expected in the viewing volume of the units?

Figure 10: the x-axis scale is unusual – please use linear or log-scale.  Also, could these results be compared against model predictions as validation of model performance?

Section 4.5:  It would be good to add a sentence or two about how the nephelometer was integrated with the DMPS for aerosol scattering coefficient distribution measurements.