h2o and co2 have lots of spikes, presumably caused by dew. ldiag is not sufficient to remove these spikes. In our QC report, we suggested using the wetness sensor, but an additional problem is that there are biases on the order of 1 g/m3 that vary throughout the project.
For CHEESEHEAD, I wrote R code to generate QC files with biases by comparison to H2O. For SAVANT, we didn't have H2O at 6m (where we have h2o), but the gradients seem pretty small, so we can synthesize a 6m from the average of 1.5m and 20m values. Indeed, a median H2O from all the H2O sensors probably would work fine. My code went on to detect spikes as large differences (say 2 g/m3) between H2O and the bias-corrected h2o. It then also applied this spike flag to co2. This code is at:
/net/isf/isfs/projects/CHEESEHEAD/ISFS/R/ec150_qc.R