In post-processing, I notice that dat(irgadiag.3m.s13) almost always is non-zero.  This was the case the entire time that this sensor was deployed at VERTEX.  I randomly looked at data from 17 Oct 00:00-01:00 and see lots of errors associated with lots of data jumps in h2o and co2.  Applying irgadiag to data up to 30 seconds prior to the flag triggering cleans up the h2o values a bunch, but still doesn't completely eliminate odd values.  Wetness shows that there may have been dew at the time (there wasn't rain).  Looking at data 12 hours later (no dew at noon), irgadiag still fires sometimes (10 times in 1 hour), though jumps in h2o are not obvious.

In the lab, I also see this problem.  S/N 1276 (the one with our internal T modification in its EC100 box), works fine.  I observed no irgadiag flags watching it for 6 minutes.  Also, the sensor head was completely quiet.  Presumably, this was deployed at s15.  S/N 1277, which presumably was at s13, makes lots of noise.  Observed for a period of about 15 minutes, there were periods of 8–99 seconds with good data, interrupted by periods of 3–11 seconds with 0x21, followed by a few 0x01 values.  These are exactly the same values, though even more frequent than, seen in the field data.  I'd say the median was 20s good, 5s bad, or only 75% good data.  I couldn't observe jumps in h2o and co2, though I was just looking at the hex output.

I note that 0x21 is supposed to indicate that "source light power exceeds limits" – an odd situation to occur.  In any case, S/N 1277 probably needs to be returned to Campbell for servicing, if only for the motor noise.

Data take-home:

1. When 0x21 values are present, the sensor is generally reporting bad values.  Filtering h2o and co2 values when irgadiag=0, i.e. the default NIDAS configuration to remove 0x21 values is insufficient to remove bad data (and sometimes appears to remove good values).
2. In general, filtering data using irgadiag is not sufficient to remove bad h2o and co2 values.  These will take more QC for (at least) dew.  I've also seen at least one s15 spike in the noon case.