Validation of A data using B data

[calvinjlin]

Channel Reliability

Currently we are using Channel A data only, but this data might be unreliable so we should compare this data to Channel B and use both channels if we can or the better of the two.

Some ways we might do this:

• Historic Data: Looking at some of the figures online, there are sensors that periodically spike to ungodly levels. These should be easy to identify and therefore exclude these channels. We can also see if the sensors start reading negative (is this possible?) or zero for extended periods of time.
• Measurement Difference in A and B: We can also look at the delta between channels A and B and if there is a sizeable gap, we can compare the channel readings to nearby PAs to see which channel on the original PA is more consistent.
• Other Sources: In a pdf produced by the EPA, it describes a method for utilizing the two A/B channels to validate each other: PurpleAir PM 2 5 US Correction and Performance ... - EPA

[HagenFritz]

Update

As per our meeting on 09/09, a few minor updates.

Comparison Interval

We want to compare the values from A and B or a short interval, but not too short to ensure we have a good overlap on measured values. The suspected optimal interval would be 10 minutes which should guarantee at least 4 data points within the same time window.

Automatic vs Manual

There are some patterns that can be be easily determined by coding them in (see Historic Data in the original issue post) while others might need to be manually identified (see Measurement Difference in A and B). The important thing is that we identify there is an issue. If we can determine which channel is problematic in the code, then we should do that. Otherwise, we should just flag the issue and figure it out manually.

[HagenFritz]

Plot generated and data removed that don't fit