Closed dlawrenncar closed 10 months ago
Note that Gretchen is hypothesizing that the absence of these high GPP values is contributing to the weaker than observed interannual GPP/Carbon cycle variability that we see in CLM5 (and to a lesser degree in prior model versions).
From email discussion:
Gordon: fpns and theta_psii are the two parameters I mentioned last week that control the light response curve for C3 plants. (C4 plants use a different relationship). fnps = 0.15, but an alternative value in some models is 0.3. theta_psii = 0.7, but in the multilayer canopy I use 0.9.
Attached is a figure from my modeling textbook, which shows the effects of these parameters on photosynthesis. I use somewhat different notation than CLM:
book = CLM ------- = ------------- theta_j = theta_psii phi_psii = 1 - fnps
The value for theta_j (theta_psii) has the biggest influence on photosynthesis. The larger value (0.9) increases photosynthesis for the same amount of PAR, but then the Rubisco-limited rate (dashed line) kicks in at lower PAR. Above this amount, photosynthesis is independent of PAR. A smaller value (0.7) has lower photosynthesis so that the Rubisco-limited rate becomes limiting at higher PAR. How this plays out in terms of canopy GPP is unclear, but it is worth trying the value for 0.9. It may be that the biggest signal is not in mean GPP but in the functional relationship of GPP with solar radiation, or even the sensitivity of GPP to direct/diffuse radiation.
From email discussion:
Danica: Yes, I think these are good parameters to add to the parameter perturbation experiment. We might also want to consider testing the sensitivity of two other aspects of photosynthesis calculations:
1) Parameters related to photosynthetic acclimation. Some of these are constants included within the equations in CLM (e.g., see lines 3166-7 in PhotosynthesisMod.F90), and other related parameters are hardcoded (e.g., see lines 2791-2807). 2) TPU. This is a highly uncertain aspect of photosynthesis (set to 1.67 * Vcmax), but note that I already wrote a paper on the sensitivity of this parameter (published in ERL, 2018).
From email discussion:
In response to question from Dave about what determines Rubisco-limited rate, Gordon replied: That's where Vcmax comes into play. It sets the dashed line. Higher values of Vcmax will give a higher light-saturated photosynthetic rate. The other parameters (fnps, theta_psii) matter when photosynthesis is not light saturated. So, too, does Jmax (but in CLM that is calculated by LUNA, not a parameter). It is hard to say how all this plays out at the canopy scale. It's likely that the sunlit canopy is light saturated, but the shaded canopy is not. But then there are a lot more shaded leaves than sunlit leaves.
Based on a paper Anthony is leading I would try changing thetaCJ and thetaIP to 0.999 (Eqn 9.6 in the tech note) this would essentially eliminate CBGB smoothing of Ac, Aj and Ap (Dave the impact will be highly analogous to your thetaJ figures above).
For diagnostic purposes:-
Also given we already did that in FATES, one could in principle look at whether FATES has the same problems, which would be interesting to know in general as well...
Agreed. I was already wondering whether or not FATES would show similar issue. We have the forcing data to be able to run at the US-UMB site if that was something that would be possible with FATES. Might be outside of the FATES climate zone where PFTs are well-parameterized, but Rosie obviously would know better about that. Something to discuss on Thursday.
On Mon, Apr 27, 2020 at 1:00 AM Rosie Fisher notifications@github.com wrote:
Also given we already did that in FATES, one could in principle look at whether FATES has the same problems, which would be interesting to know in general as well...
— You are receiving this because you authored the thread. Reply to this email directly, view it on GitHub https://github.com/ESCOMP/CTSM/issues/993#issuecomment-619773003, or unsubscribe https://github.com/notifications/unsubscribe-auth/AFABYVBGZJA35POGEESJ6XLROUUSRANCNFSM4MQL7BBA .
Also given we already did that in FATES, one could in principle look at whether FATES has the same problems, which would be interesting to know in general as well...
I was thinking it would also be interesting to look at these relationships at different sites with differences in vegetation structural diversity. We might expect a more mixed response to diffuse fraction in a mixed-age, mixed-species stand like at UMBS because different shaped tree crowns are better able to absorb different levels of diffuse/direct light, whereas in a more even-aged, low species diversity stand we might expect a stronger relationship with diffuse fraction.
We could try to leverage the various NEON sites which have both tower and lidar/forest inventory data and compare to FATES?
Thanks @alistairrogers, @xuchongang, @walkeranthonyp, @rosiealice, @danicalombardozzi for your comments and suggestions today. I also found this discussion in FATES helpful & similar to part of our conversation today. I'm going to summarize suggestions from the meeting here. Please modify or add to this thread as is warranted.
Ideas to test, not necessarily solutions. A goal is to try to understand where the limitation is sourcing from LUE (LUNA parameters), CBGB scaling, light saturated photosynthesis, or canopy scaling (including light profiles through the canopy).
[ ] Double Vcmax (LUNA). https://github.com/ESCOMP/CTSM/blob/3c31fb41bf55de43e71637aaa0bd7c34752332f1/src/biogeophys/LunaMod.F90#L75
https://github.com/ESCOMP/CTSM/blob/3c31fb41bf55de43e71637aaa0bd7c34752332f1/src/biogeophys/LunaMod.F90#L76
This will increase both Jmax and Vc,max.
Note For Arctic bugs #990 & #961, Jmaxb1
was these lowered for CLM5 for testing from 0.17 in the Ali (2016) paper.
[ ] Jmax:Vcmax ratio
This is tc,j0
, from Ali (2016)and this parameter in the code https://github.com/ESCOMP/CTSM/blob/3c31fb41bf55de43e71637aaa0bd7c34752332f1/src/biogeophys/LunaMod.F90#L77
[ ] CBGB scaling (Collatz), Smoothing parameterization (including TPU) may be important.
theta_ip = 0.95_r8
& theta_cj(p) = 0.98_r8
NOTE theta_cj is different for C3/C4. @alistairrogers suggested running with both of these set higher (0.999)
[ ] Leaf level light response of photosynthesis: (still have some light saturation, hard cap on leaf-level photosynthesis
fnps = 0.15_r8
& theta_psii = 0.7_r8
Try increasing theta_psii = 0.9
, seems like fnps may already be high enough?
[ ] CANOPY SCALE
Leaf orientation and angle, see new parameter files (leaves too flat, (chi_)
pftcon%xl
used in SurfaceAlbedoMod
For xl, -1 means vertical leaf, 1 is horizontal leaf, 0 is random orientation. Note that new optical properties in issue #807 move xl values towards higher (closer to 1) for most PFTs.
Thanks Will and Dave for invitation to join the conversation. I'll be interested to stay in touch and hear what you learn.
I suspect changing Theta_psii will not have a large effect so maybe put that one lower down your list. Theta_psii at 0.7 seems reasonable to me and is widely used in many TBMs. It allows a gradual transition from light limited to carboxylation limited photosynthesis. Cranking it up to 0.9 probably won't help much as your low GPP is at saturating irradiance rather than in the transition zone between light limited and light saturated photosynthesis.
In my mind all signs point to low carboxylation capacity as the culprit.
We will include it. It's actually #1 in list, which should read 'double Vcmax, rather than just Vcmax. One reason, though, that we are not as focused on this is that it looks like CLM prediction of Vcmax from LUNA is not 'too bad'. See Table 3 in CLM5 overview paper. https://agupubs.onlinelibrary.wiley.com/doi/epdf/10.1029/2018MS001583%4010.1002/%28ISSN%291942-2466.CESM2
But, either way at this point, we are just exploring sensitivity, so it is worth a try.
On Fri, May 1, 2020 at 1:33 PM alistairrogers notifications@github.com wrote:
Thanks Will and Dave for invitation to join the conversation. I'll be interested to stay in touch and hear what you learn.
I suspect changing Theta_psii will not have a large effect so maybe put that one lower down your list. Theta_psii at 0.7 seems reasonable to me and is widely used in many TBMs. It allows a gradual transition from light limited to carboxylation limited photosynthesis. Cranking it up to 0.9 probably won't help much as your low GPP is at saturating irradiance rather than in the transition zone between light limited and light saturated photosynthesis.
Also don't forget to try doubling Vcmax (not on your list immediately above Will). It would be easy to implement and in my mind all signs point to low carboxylation capacity as the culprit.
— You are receiving this because you authored the thread. Reply to this email directly, view it on GitHub https://github.com/ESCOMP/CTSM/issues/993#issuecomment-622530864, or unsubscribe https://github.com/notifications/unsubscribe-auth/AFABYVFY2UJ6N2P6O4GR533RPMPZRANCNFSM4MQL7BBA .
Seconding @alistairrogers - it was great to chat with you about this. Thanks for the invitation! I hope it's useful in the end and looking forward to seeing the results of your investigations.
Also to add (as I didn't see it in your list @wwieder) that it'd be really useful to see GPP plotted against APAR.
I agree with @alistairrogers that the problem seems to be canopy-scale carboxylation. One thing to note, however, is that the calculation of Vcmax is quite different between CLM4.5 and CLM5, and we were seeing similar caps to photosynthesis in CLM4.5. This also leads me to wonder about the influence of canopy scaling, as I'm not sure whether that has changed as much. It's worth working through each of the steps that @wwieder outlined above.
pftcon%xl
used in SurfaceAlbedoMod For xl, -1 means vertical leaf, 1 is horizontal leaf, 0 is random orientation. Note that new optical properties in issue #807 move xl values towards higher (closer to 1) for most PFTs.
I'm probably missing something but I'm not used to this way of expressing xl
given that when leaf angle is translated into an extinction coefficient, random leaf angle distribution is 0.5, horizontal 1, and vertical 0. In any case, one way to hack clumping into the canopy if you don't have a separate parameter would be to multiply the extinction coefficient by a clumping factor 0-1 where 1 is no clumping and 0 infinite clumping (i.e. zero light interception). That could be done through xl
but I'm not sure how xl
should be adjusted, reduce it to so that xl < 0
I guess.
https://www.geosci-model-dev-discuss.net/gmd-2019-335/gmd-2019-335.pdf
This paper is also very relevant. Different ecosystem (Costa Rican forests) but similar(ish) results. See Fig 4 especially. They find that saturation GPP is good but that the slope is too high and they correct by increasing fnps
(if I'm getting my parameter names correct) so that less absorbed light ends up stimulating electron transport.
So it's possible there are two things at play: 1) the saturation point is too low (in the UMBS and other sites, but not CR, where else?), and 2) the curvature of the light response may be too strong.
@dlawrenncar thanks for the reference. Is this your preferred citation for CLM5?
I see that your Vcmax25s largely match K2009. The problem with CBGB smoothing is that the K2009 Vcmax25s are fit from datasets that did not include smoothing in their fitting approach. To achieve the same A as a model without CBGB smoothing would require a larger Vcmax.
Yes. That is the standard reference for CLM5, though depending on topic, there are a lot of other CLM5 papers out there (Rosie, Gordon, Will, Danica, Daniel all have papers covering various related topics!)
On Fri, May 1, 2020 at 5:08 PM alistairrogers notifications@github.com wrote:
@dlawrenncar https://github.com/dlawrenncar thanks for the reference. Is this your preferred citation for CLM5?
I see that your Vcmax25s largely match K2009. The problem with CBGB smoothing is that the K2009 Vcmax25s are fit from datasets that did not include smoothing in their fitting approach. To achieve the same A as a model without CBGB smoothing would require a larger Vcmax.
— You are receiving this because you were mentioned. Reply to this email directly, view it on GitHub https://github.com/ESCOMP/CTSM/issues/993#issuecomment-622604904, or unsubscribe https://github.com/notifications/unsubscribe-auth/AFABYVGJPNODBX6DCMRV6OTRPNI7DANCNFSM4MQL7BBA .
@alistairrogers , in your suggestion to double Vc,max I think this will also increase Jmax. This this OK? @xuchongang is there any way to modify only Vcmax (or Jmax) alone, or is this just done by changing Wc2Wjb0
, which controls their ratio?
For what its worth, NEON includes data on PAR as well as direct and diffuse fractions, allowing us to test this at multiple sites down the road!
To summarize, here's the list of parameters we're focusing on:
I wonder if it would make sense to relax the limit on wc2wjb0 in any next set of parameter perturbations. Chongang implied that it may not be necessary or even appropriate.
I was really struggling to map the LUNA parameters to traditional FvCB parameters on the fly. If I understood it correctly wc2wjb0 controls the JVratio. See fig 5(c) in Kumarathunge et al (2019) for the global range of JVratio25. Checking that the JVratio25 is close to that predicted for the growth temperature makes sense.
Agreed, it would be good to check site values of Vcmax and Jmax (JV), and try to run CLM without LUNA and the site values of V and J. If that's possible.
Also, this may be obvious, sorry if so. If I remember right high JV was decreasing the GPP cap. You wouldn't expect that if V and J 25 were simply fixed parameters. But LUNA is trying to optimise a fixed quantity of N, so higher JV means a higher fraction of N in J per unit V. Assuming the amount of leaf N is fixed (which it isn't i guess in CLM5.0, but for simplicity), then that means higher JV results in lower V, and therefore a lower cap given the cap is V limited (light saturated photosynthesis).
@wwieder Hi Will could you point me to the relevant bit of code or confirm the units of Ncb in the code. (see Eqn 2.10.2 in the tech note and the legend for Table 2.10.1) Ncb is listed as having units of gN associated with Rubisco m-2 leaf area and also as gN in Rubisco g-1 N in the leaf. I'm in the middle of a related conversation with Jenifer Holm and this could shed some light on this issue. There's a potential -23% error in Vcmax25 but it might just be an error in the tech note.
The code is labeled similarly https://github.com/ESCOMP/CTSM/blob/b0547837413ee9fe5e8039da5814882b4faf4cbd/src/biogeophys/LunaMod.F90#L874
Here's where it's calculated https://github.com/ESCOMP/CTSM/blob/b0547837413ee9fe5e8039da5814882b4faf4cbd/src/biogeophys/LunaMod.F90#L919
Thanks Will, The CLM5 tech note lists PFT specific parameters for Ncb that are equal to the FLNR values for these PFTs in CLM4.5 (Jenifer spotted that one) but the units are different. The units in the code are consistent with calculating Vcmax but if you use the values for Ncb in table 2.10.1 in the tech note you get a lower Vcmax because you haven't yet multiplied by Na. This wouldn't be obvious for Na close to 1 but for BDT-temperate you would underestimate Vcmax25, that would limit CO2 assimilation at high irradiance. It would be useful to see PFT specific values for FNCa and PNcbold to see if the tech note matches the code.
Had some more time to chase parameters around in the code. Old Thorton & Zimmerman equation from CLM4.5 Vcmax = Na FLNR FNR * aR25
LUNA equation (line 444) Vcmax25_opt = PNcbopt FNCa Fc25
Fc25 maps to FNR*aR25 FLNR maps to PNcbopt and PNcbold FNCa maps approximately to Na
The code seems sound but the technote is very misleading, specifically the equations and table mentioned above.
In CLM5 and prior versions, maximum hourly values of GPP are not as high as in observations at at least some tower sites, definitely at US-UMB site. See attached figures from Gretchen Keppel-Aleks.