search

forc-db / Global_Productivity

Creative Commons Attribution 4.0 International
2 stars 0 forks source link

ANPP/NPP measures #33

Closed beckybanbury closed 4 years ago

beckybanbury commented 5 years ago

@teixeirak I'm just trying to finally pin down which NPP and ANPP variables I'm using, and make sure we're completely happy with our approach. I think for NPP it's best to use just NPP_1 (including foliage, branch, stem, coarse root and fine root). For ANPP, for the general analysis, I'm thinking to include ANPP_1 (foliage and stem) and ANPP_2 (foliage, branch and stem) in order to keep it consistent with NPP. I had just been using ANPP_1, but I don't want to be using NPP measures that include branch, while not including that in the ANPP measures. What do you think?

The only thing to note with using both ANPP_1 and ANPP_2, is that we are not using ANPP_woody, so our component fluxes do not include any measure of branch productivity. Unfortunately we don't have a measure of NPP which excludes branch productivity.

For some of the more specific analyses e.g. the stacked plots, I have used the terms which are nested within each other, to keep it consistent (e.g. NPP_1 = BNPP_root + ANPP_2)

teixeirak commented 5 years ago

Hmmm, that's tricky. Could you include variable type (wether branches are included or not) as a covariate in the analysis?

teixeirak commented 5 years ago

For the record, ANPP_woody_branch is usually less than one and doesn't appear to vary much with latitude.

beckybanbury commented 5 years ago

Hmmm, that's tricky. Could you include variable type (wether branches are included or not) as a covariate in the analysis?

I can do that - I tried running it as a random effect; however it doesn't improve the model at all (the best models are coming out as those without that term), so I guess that the effect of woody branch is sufficiently small that there is very little impact on the model.

teixeirak commented 5 years ago

Should it be a random or fixed effect? I almost want to force it to be in the model, but if it really makes no difference we could combine the two (favoring records with branch when available), and state in the methods that it didn't make any difference. In theory it should matter, but in practice that may be swamped by other sources of variation.

beckybanbury commented 5 years ago

I included it as random, because we would expect to see correlations within the two levels of group (ANPP_1 and ANPP_2), but differences between them. Random effects account for the variation between levels that isn't correlated with the fixed variables; I think you most often use them where you have data in groups or levels (and fixed effects tend to be continuous variables). Please correct me if you think I'm wrong though!

I can include it in the model, but I think it will affect the outputs when we're looking at the multivariate models, because it adds an extra term without explaining additional variation. From my initial analysis including it as a term, that means that AIC analysis will penalise models with more terms, so we may get less reliable results from the multivariate analysis. We'd have to compare models with and without variable.name as a term to get round this.

teixeirak commented 5 years ago

Just leave it out then (combining data with and without branch productivity).