Next steps

[bpbond]

After talking with @wilsonsj100 and @megonigalp today.

Straightforward

• [x] Implement default dead band time and custom dead band (by observation in metadata)
• [x] Problem with CO2? flag associated methane observation
• [x] Polynomial model - at least fit and use R2 divergence from linear as a diagnostic
• [x] y-intercept as useful QA/QC

Trickier

• [ ] Flag observations/fluxes below detection limit
• [ ] Identifying ebullition events - how? - possible before/after slope changes <-- a segmented model?

[wilsonsj100]

I think we should also consider adding into the beginning an addition of 10-30 secs to the start time as an equilibration time - i.e. to ensure equilibration of the tubing and closed chamber. This is another little things so not a big deal.

[bpbond]

@wilsonsj100 I've set up a standard QA/QC "compare two things" framework for this. Basically, first we plot column 1 versus column 2 and use standardized residuals to identify outliers. For example:

Then, we plot the potentially problematic observations in detail:

Feedback welcome!

[bpbond]

Nice, the polynomial test is correctly identifying a curvature problem in one observation:

[bpbond]

Superseded by #11