Define statistical model

[teixeirak]

I'm out of time to do this now, but this is a reminder.

[teixeirak]

We'll use a nonlinear mixed effects model.

Dependent variables

We'll want to try a number, and of course this will be easy once we have the code. To start:

• GPP
• NPP
• ANPP
• ANPP_woody

Fixed effects

To start:

• MAT OR |latitude| (separate analyses- one with each)- this is the nonlinear term. Let's use with a 2nd order polynomial fit.
• leaf type (broadleaf / needleleaf / mixed. Combine evergreen/deciduous categories for now)

Potentially later:

• age (we will want to look at this one, but I'd like to work out the relationships in mature forests first)
• MAP
• leaf phenology (evergreen/deciduous)

Random effects

• plot nested within geographic.area

Factors to control for

• [ ] age >=100 (or 999)
  • We may add this as a fixed effect later.
• [ ] management/ disturbance
  • remove stands that are too strongly influence by management/ disturbance. relates to this issue.
• [ ] Methods
  • Required covariariates:
  • min.dbh
    • remove any with min.dbh >10cm
    • [assess effect from min.dbh = 0 to min.dbh = 10]
  • allometry (?) - This can matter, but we won't address it for now.
  • general method [these should be assessed]
  • GPP
    • NEE + ecosystem respiration
    • NPP + autotrophic respiration
  • NPP
    • Different variables

[ValentineHerr]

@teixeirak, should I combine all ANPP (ANPP_0, ANPP_1, ANNP_2..)? Should have I done that when creating ForC_simplified?

[teixeirak]

This is a very good question! It feeds into the larger question of how finely we try to dissect methodological differences. Let's combine for now. At some point we should test whether there's a significant effect of treating variable id as a random effect. (@nkunert, I'd be happy to get your comments on this)

They should NOT be combined in ForC_simplified.

[nkunert]

@teixeirak, @ValentineHerr, I need some more details on this. Does combining "ANPP_0, ANPP_1, ANNP_2.." mean that you average ANPP from different plots regardless of how ANPP was measured/estimated? ANPP derived with different methods for a given forest ecosystem should match up pretty well, thus for looking at ANPP on a broader ecosystem perspective, it is fine. However, a simple table separating ANPP into different methodological approaches could be a nice thing to have - potentially there is a reviewer out there who wants to see this!

[teixeirak]

@nkunert, details on these variables are here. Basically, the difference is in whether they include branch turnover (and whether we have info about this). There are lots of other examples of methodological differences of similar magnitude that we'll need to consider.

[ValentineHerr]

@teixeirak , some stand.age are 0. Is it OK to consider that a forest? It is technically only seedlings, right? Either way, I can't take the log of 0 for the analysis. Should I just add a tiny number to make it different than 0 or should we handle that differently?

[teixeirak]

Assuming there are very few, let’s just drop these.