search

PNNL-TES / GlobalC

A topdown and bottom up approach for global C cycle
6 stars 4 forks source link

Breaking down the GPP estimates into more categories #68

Open bpbond opened 4 years ago

bpbond commented 4 years ago

@ashiklom comment in Google doc:

This actually gets back to the idea of breaking down the GPP estimates into more categories -- the ones that are based on satellite products vs. ones based on flux towers vs. truly "top-down" approaches based on atmospheric inversion or whatever. If those are systematically different, that may be an important part of the narrative here. If not, we can mention it briefly in the text and point to a supplementary figure.

jinshijian commented 4 years ago

Hum, it looks like top-down GPP (based on FLUXNET, O18, O13 etc) are lower than bottom-up GPP and process-based model (e.g., ESMs) GPP, please see the results below, image

bpbond commented 4 years ago

Huh! That's interesting.

bpbond commented 4 years ago

Just checking, SIF is in the top-down category here?

ashiklom commented 4 years ago

Echoing Ben's comment above, it would be useful to know exactly what goes into each category. Personally, I would consider both FLUXNET and SIF estimates as "bottom-up", since they are averages of site-level fluxes (...right?), whereas the isotope analyses are probably more "top-down".

You might also consider other groupings -- e.g. remote sensing vs. isotope vs. tower.

Another relatively easy and possibly informative figure would be to plot GPP estimate (y) ~ method (X), with X sorted by ascending/descending GPP and see if any clusters jump out.

jinshijian commented 4 years ago

Hum, good idea. SIF is in the top-down here now, should change that to bottom-up? Good idea to mention what going to top-down and what going to bottom-up. I will make a note in the google doc. Thanks

jinshijian commented 4 years ago

I grouped GPP into three groups: Bottom-up: Fluxnet, isotope, SIF Top-down: Satellite Mixed: C13 with model, satellite with Fluxnet, satellite with LAI data

I try to do this" Another relatively easy and possibly informative figure would be to plot GPP estimate (y) ~ method (X), with X sorted by ascending/descending GPP and see if any clusters jump out." but don't know how to sort X by ascending GPP, can you find some example online and share it with me @ashiklom ? Thanks See figure attached image