So far we have seen how distance and geography can impact the linear fit models between PurpleAir and AirNow. From our tests in the Methow Valley, it became clear that there is no universal scale factor between the two services since the models changed considerably the further apart the monitors were located. What we were able to determine though, from the locations we tested, is that PurpleAir units tend to measure roughly twice as high as those from AirNow.

In this report now we’ll be exploring how these models change not over distance, but over time. How do the scale factors change? By how much? What about their overall R-Squared fit? We are interested to see if the models for individual PurpleAir monitors remain relatively unchanged day-to-day, week-to-week, and so forth. Even if the models must change due to distace, hopefully they do not vary considerably from one interval to the next.

Methods

Since we’ll be following the same instructions for generating week-by-week AirNow and PurpleAir plots, both fitted and unfitted, here is a brief outline of the tasks performed for each week “snapshot”.

  1. Load the PurpleAir and AirNow hourly timeseries data for the desired location using PWFSLSmoke, MazamaPurpleAir, and openair package functions.

  2. Geneate a linear fit model from PurpleAir to AirNow with R’s built-in lm() function.

  3. Plot both “fitted” and “unfitted” PurpleAir timeseries data on top of AirNow readings using R’s base plot functions.

Winthrop

We’ll begin again in the Methow Valley, using the Winthrop Library PurpleAir unit because it is the closest to the downtown AirNow monitor.

Monitor names:

Here we see that, between 3 weeks, the scale factor for the Winthrop Library model does vary but stays roughly around 0.6. We can assume then that the general rule of thumb for downtown Winthrop is to halve PurpleAir values to get close to the AirNow readings.

Twisp

Further down the Methow Valley, in Twisp, we’ll check out the town hall PurpleAir monitor and compare it to the AirNow monitor just across the street.

Monitors used:

Much like in Winthrop, here we also see that the scale factor for the Twisp Town Hall monitor changes week-to-week. Again, the factor hovers around 0.5, meaning that the Twisp Town Hall PurpleAir unit reads twice as high as as the Twisp AirNow monitor.

Yreka

So far we’ve seen that the PurpleAir monitors in the Methow Valley towns generate different scale factors over time, but they are all relatively close to 0.5. But what about different locations? Let’s see if other regions agree with this factor by taking a look at Yreka, CA, where there are two colocated monitors.

Monitors used:

Now we have something a bit different. The scale factors in Yreka tend to measure somewhere around 0.6 or 0.7, meaning that the PurpleAir readings are actually closer to the AirNow values than in the Methow valley towns. You don’t have to reduce the PurpleAir data by a whole 50% here, and this is true not just for one week, but several, though the exact scale factor changes slightly.