walkeranthonyp
commented
3 years ago There's a further hybrid option of those you list above, but it's going to add complexity.
- Hybrid option(s): Different dinc_ed for RT and for calculation of photosynthesis. e.g. calculating RT with dinc_ed of 0.2 or smaller, and photosynthesis with dinc_d of 1 ... or perhaps variable by canopy layer. If RT calculation is still too expensive Gordon's textbook has a tridiagonal solver for the Norman scheme (tho I recognise that's not trivial to add with with this awesome fates RT code!).
By two-stream do you mean the same as the two bigleaf approach with a single sun and shade leaf (as in clm), but I guess one sun n shade leaf per canopy layer (per pft)? If so I guess option 4 is a slightly more complicated version of that.
ckoven
rosiealice
Looking at the SP mode simulations, it appears that FATES has a low bias in the absorbed radiation fluxes (and a correspondingly high bias in reflected radiation), compared to CLM5-SP runs with the same LAI forcing, no stem area index (for the sake of avoiding #744) and optical properties.
The dinc_ed parameter is hard-wired, here: https://github.com/NGEET/fates/blob/516c1845d5fe18eacbb7e8e7d70596b4eff364aa/main/EDTypesMod.F90#L123
Modifying it from 1.0 to 0.2 reduces the bias from ˜0.15 (version 4 below) to <0.06 (version 5 below).
Obviously, this is non-ideal. The potential fixes include:
'1' is just really for the sake of completeness and/or testing. Making dinc uniformly <0.2 through the canopy would be far too expensive.
'2' is less of a time sink than '3'., and is probably what we need to do in the interim. '3' would allow one to replicate 'big-leaf' physiology in the FATES code and thus ultimately reduce duplication.
˜