Field calibrations of a low-cost aerosol sensor

Our  paper on the development and evaluation of a PPD42NS-based instrument is now publicly available via the AMT Discussion forum (Article [PDF]Supplement [PDF]). It’s permanently citable in its existing form, though it will technically be in review until March 24. Hope you find the results to be of interest, dear reader.

PPD42NS sensors on a calibration rack

6 thoughts on “Field calibrations of a low-cost aerosol sensor

  1. Hi David,
    your paper on AMTD is really very interesting. Time ago I’ve tried to test the PPD42 in our lab, but our results were not so good.
    We used the same Arduino configuration and 60 sec average. The problem is that by most of the time we get zero, and some sporadic “spikes”. Hourly average seems to have some correlation with reference instrument, but I’m concerned with the long sequences of zero. ( we measured up to 1h consecutive 0)
    I have to say that your average concentrations are about twice the concentration we measure. I mean we have much lower dust concentration and this could influence our results.
    I’m wondering if there is some mistakes in our setup, and maybe you can give us some hints, or confirm that you observe the same behaviour .
    Any help will be appreciate.
    Thank you and ciao

  2. Hi Fabrizio, thanks for the comment. So far we’ve not attempted calibrations on timescales smaller than 1 h.

    In our experience, it’s true that the PPD42NS can run for many minutes in lab/office air without emitting a pulse, and it has seemed that longer runs of consecutive zeros are more common in regimes with a lower aerosol concentration. This would be consistent with the working hypothesis that the time-weighted average (TWA) correlates with some measure of the aerosol concentration. I’m not surprised that one (?) of your deployed sensors went at least once for a full hour without emitting a pulse when conditions were between 1-10 µg/m3 PM2.5 (?). I’d be surprised if this were very common under conditions between, say, 10-100 µg/m3 PM2.5.

    Have you published anything about your setup?

  3. Hi David,

    Do you have the code available for measuring the Shinyei’s signal as described in your paper? I’m looking for how you captured the signal at 1 Hz.


    • Hi Anthony, thanks for the comment. It’s great that you are challenging the notion that there is not a temperature effect. Almost surely there is—these are electronics, after all. I would say that what matters is whether the effect is negligible and, if not, the conditions under which it is manifested.

      On the chart you linked to, the range of variation in % FS (aka % LPO) is reasonably large. If all or most of it is due to T, that would be important to take into account. It does look like there is a strong inverse relationship in those data. The temperature range (10-15º) (Celsius?) is outside the range of temperatures in our dataset (approx 20-30 ºC). That might result in a different relationship between measured temperature and measured % FS. Possibly for unexpected reasons.

      I can’t comment much more deeply on the chart, since I don’t know enough about the circumstances under which those data were collected. (I wish I had more time to follow up but I’m crushed with many obligations at the moment!) If those are observational data, one or more lurking variables might explain some, all, or none of the covariance.

      We claimed only that we failed to observe a significant correlation between T and % FS within the scope of our dataset. If we had reason to believe that there was a causal relationship—perhaps one that was masked by other variation—I would be glad to know it and to go back and adjust estimates accordingly.

      The best thing would be for someone who has time to set up a controlled experiment and manipulate temperatures carefully on the experimental arm. We’d be very interested in the results.