running issues with SRER and F001,010,001

[jmzobitz]

Clear there is an anticorrelation with F_{010}. Figuring out why. Stay tuned.

[jmzobitz]

Also #9 talks about this.

[jmzobitz]

CO2:

Diffusivity:

[jmzobitz]

Ooops - forgot to add in the surface CO2 from linear regression:

OK! I think I understand why. @naupaka: Chew on this. :-)

• F[010] is derived from CO2 at -0.02 and -0.06.
• F[001] is derived from CO2 at -0.06 and -0.18.
• F[100] is derived from CO2 at 0 and -0.02. The problem is that the measurement at 0 (purple dots) is a linear regression across all three measurement layers, which may be unrealistic because there is bigger drawdown of CO2 near the surface - whereas the layers at -0.02 and -0.06 seem more compressed --> which make sense that the F[010] is tending towards 0 and then increases.

I think that the linear extrapolation to the surface may be an issue and introducing larger gaps. So perhaps strike F[100] as a valid method?

Figured this out just as soccer was ending, so yay!

[jmzobitz]

Removing F[100] and swapping F[111] to F[011] changes everything: 🤯 (and I mean in a good way)

[jmzobitz]

Discussion: 3 distinct with common scenarios, standard assumptions not necessarily met - what this analysis shows, even in those case, if you are aware of those cases, you get pretty darn close with an alternate method. Case of WREF (saturdated soil), KONZ (rich soil with shallow pulse and re-wetting event), SRER (super hot soil that cools at night) Can't apply a single method across all sites / conditions.

Assumptions: CO2 concentration rate of change is constant with depth- r2 boxplot of regression with F00 and F11 methods.

[jmzobitz]

@naupaka: Some follow ups from yesterday:

• I changed my method to adjust the Expanded Uncertainty so it reflects 1 SD (per Ed's suggestion)
• I liked your idea of computing the fluxes across layer 1 and layer 3 and ran with it.
• After reflecting on the struggles it took to articulate differences between F[11], F[10], F[01], and F[00], I am changing things up.
• F[ijk] refers to which measurement levels were used to compute dC/dz in the flux gradient method. So F[011] - deepest two levels, F[101] - first and third level.
• F[111] no longer exists - I think it was suffering from the same issues that extrapolation of the CO2 measurements to the surface were (I can explain on Tuesday more).

The upshot is I think this simplifies the methods used (and also in the storytelling as we write this up). We are only computing fluxes using instruments provided by NEON.

I am pretty jazzed by these results:

Here is what I would say:

• SRER findings (sandy soil, driven more by the activity in the upper two layers) is consistent. F[000] and F[101] are influenced by the deeper layer, which is lagged in terms of activity from the upper two.
• SJER: this site has the most gap-filled data but perhaps F[000] is a good substitute because proxy in times such as these?
• I can develop a test to look at time points where flux gradient assumptions are valid - I am thinking it will be something like F[110] > 0 & F[011] > 0 & F[101]>0 (maybe not the last one?), but that should take care of conditions when CO2 increases through the depth profile. Unsure if I want F[110] < F[011].

[jmzobitz]

Closed, as we know what is going on at the different sites.