Increasing LAI in default run

[rosiealice]

Here is a record of the discussion of low LAI we have been having on an email thread.

@kovenock Abby and I are looking for ways to simulate higher LAI ecosystems in our Fates simulations at bci. Currently, we are using the parameter file called fates_params_2troppftclones.c171018.nc. With this parameter set, the annual mean LAI is around 2.2 m2/m2 at 400ppm CO2 (and 2.4 m2/m2 at 800ppm CO2). We think that the initial LAI level could alter the outcome of our competition experiments and would like to test the influence of starting from a higher LAI ecosystem.

Do you have suggestions on how to best increase the LAI at bci in Fates simulations?

Jennifer Holmes has mentioned to me that she has a parameter file for the gf2 site that gives a higher LAI ( ~ 6 or 7 m2/m2 if I remember correctly) but I thought I would check with you first for suggestions and/or advice on what is important to consider.

@ckoven We were just talking about this on our recent modeling call. There are a few issues leading to the low LAI bias in current parameter sets. One is that we don’t have a healthy understory. I’m about to submit a PR that reduces respriation rates when a plants carbon stores are low (currently they die at a faster rate, presumably due to deferred maintenance costs, but they don’t actually defer those maintenance costs, which my PR changes). I’ve tested this and confirmed that it does produce an understory. A second issue is that, in earlier attempts to have an understory, we specifically reduced the allometric coefficient relating leaf biomass and stem diameter so that a given tree would have a thinner canopy. The idea was to let light into the understory, but in practive it didn’t work very well, and the net effect was to put a community-level LAI cap on the forest. So we need to increase this parameter again.

Rosie was going to do some investigations of whether the canopy trimming logic was also involved in this, but I’m not sure the status of that.

I’m going to try to kick off some simulations using my respiration-throttling code later today, and will also try a case where I increase the leaf biomass : DBH coefficient. Hopefully some combination of those will generate decent LAI.

@rosiealice I have done some experiments increasing the canopy LAI using the minimum canopy spread parameter. Figures attached. In summaryL

1) Decreasing the minimum spread (allom_d2ca_coefficient_min) from 0.33 to 0.10 increases LAI from ~2.4 to ~3.5 in the Amazon

2) In doing so, we appear to turn on the 'trimming' algorithm.

3) The understory biomass is a) lower in the new parameetization, as expected, and is also significantly more chaotic.

My feeling is that going from being limited by the maximum allometry to being limited by trimming pushes the model into some kind of alternative state, and ultimately, we do ned to improve the light getting to the understory to make things better.

@rgknox Another thing we may want to try is modifying the optical parameters, like backscattering coefficients, and clumping. If the system is not nutrient or water limited, its light limited. We may be able to see if the carbon balance is as efficient as it should be, which could help us increase LAI, but otherwise, I would guess that we need more light penetration per unit LAI.

@ckoven I still think that of the two separate ways to achieve high LAI of the canopy layer [which are (1) make narrower cylinders with the same leaf biomass:DBH relationship, vs (2) increase the leaf biomass: DBH coefficient], we probably want to be doing more of (2) than (1). I say that because in principle (1) will tend towards much higher stand densities and therefore biomass. But I believe that neither our biomass, nor our stand densities are as wrong as our LAI…

@rosiealice Agreed. Also, however, if we are invoking trimming at LAI ~3.5 then we'll need to do something to the carbon economics of the lower leaves to get any higher than that as well as increasing the maximum allometric LAI?

[ckoven]

@rosiealice to your last point, that's right. so a few possibilities occur to me:

• turn off trimming entirely (I'm running a case like this just for curiosity right now)
• reduce understory mortality rates by respiration throttling (note that the intersection of the understory respiration throttling logic that I've recently added and the leaf trimming logic is possibly unstable)
• allow greater light penetration into the canopy, through some combination of:
  • Imperfect plasticity (possibly somewhat involved, and in any case only allows leaves in the understory to be more in positive carbon balance, still limiting a max LAI depth of the actual canopy layer)
  • leaf clumping; looking at the code, it seems like this is possibly as simple as just multiplying by a clumping factor for the kdir term in https://github.com/NGEET/fates/blob/master/biogeophys/EDSurfaceAlbedoMod.F90#L254 (I just checked with @rgknox who doesn't think the above statement is crazy)
• play with the leaf photosynthetic parameters to increase the relative assimilation versus respiration (the numbers in our parameter file right now are pretty random, so what we really ought to do on this front is make sure these correspond to observations; I don't want to invoke the wrath of @alistairrogers).

I'm sure other possibilities exist too. Which of these seem most promising to people?

[mdietze]

I think @ckoven's list of possibilities correspond to a set of hypotheses about what's wrong. Rather than picking the "most promising", I'd ask what corroborating observations support each of these (i.e. each would make different predictions about other quantities that we have data on, such as understory light conditions, understory demographic data, leaf trait data, etc).

[rosiealice]

I'd add to all of the above some perturbation of leaf economics parameters, since those affect both the allometric and carbon balance /trimming based maxima...

On Mar 1, 2018 8:04 AM, "Michael Dietze" notifications@github.com wrote:

I think @ckoven https://github.com/ckoven's list of possibilities correspond to a set of hypotheses about what's wrong. Rather than picking the "most promising", I'd ask what corroborating observations support each of these (i.e. each would make different predictions about other quantities that we have data on, such as understory light conditions, understory demographic data, leaf trait data, etc).

— You are receiving this because you were mentioned. Reply to this email directly, view it on GitHub https://github.com/NGEET/fates/issues/339#issuecomment-369619925, or mute the thread https://github.com/notifications/unsubscribe-auth/AMWsQ5e_NIDXE4fNsbDInK2u3X5jq8D-ks5taA3hgaJpZM4SXP8C .

[serbinsh]

Agree with @mdietze @rosiealice and others regarding a set of data-informed exercises. In addition, the leaf trait data, photosynthesis data, etc that has been collected for NGEE et al., we have good data on leaf optical properties we can leverage as well; as mentioned above regarding leaf single-scattering albedo (i.e. leaf refl+trans) from measurements in NGEE and elsewhere.

For example as we have shown with MAAT our measurements of photo params from Panama (both ENSO and past work by @walkeranthonyp et al) can inform/constrain diurnal and seasonal assimilation. We can easily provide updated params to start but long term carry out the UQ work we have been discussing to help guide this with MAAT && ultimately FATES.

This is one reason I wanted to get the site inventory runs working so we can invoke these within PEcAn etc to test different constraints we have for some of these params absent spin up.....Dont want to step on toes so I think there is a lot of parallel work that can happy on this.

[serbinsh]

Also, FYI - as we add data to the PEcAn db you can also start playing with observations for FATES. For example

#**Extract data**
tropical_Vcmax <- betydb_search("Lianhong Gu Vcmax", include_unchecked=TRUE)
    tropical_Vcmax %>%  group_by(scientificname, trait) %>%  
    mutate(.mean = as.numeric(mean)) %>%  
    summarise(mean = round(mean(.mean, na.rm = TRUE), 2), 
        min = round(min(.mean, na.rm = TRUE), 2),
        max = round(max(.mean, na.rm = TRUE), 2),
        n = length(n))

#**Merge data (by entity field)**
merged_data <- merge(tropical_Vcmax[,c("scientificname","entity","mean")],
                tropical_Jmax[,c("entity","mean")],by="entity")

[rosiealice]

Vcmax somewhere between 15 and 80 then! Think we'll probably just about manage not to get that wrong ;)

On Mar 1, 2018 9:04 AM, "Shawn P. Serbin" notifications@github.com wrote:

Also, FYI - as we add data to the PEcAn db you can also start playing with observations for FATES. For example

Extract data

tropical_Vcmax <- betydb_search("Lianhong Gu Vcmax", include_unchecked=TRUE) tropical_Vcmax %>% group_by(scientificname, trait) %>% mutate(.mean = as.numeric(mean)) %>% summarise(mean = round(mean(.mean, na.rm = TRUE), 2), min = round(min(.mean, na.rm = TRUE), 2), max = round(max(.mean, na.rm = TRUE), 2), n = length(n))

Merge data (by entity field)

merged_data <- merge(tropical_Vcmax[,c("scientificname","entity","mean")], tropical_Jmax[,c("entity","mean")],by="entity")

[image: image] https://user-images.githubusercontent.com/3067369/36854907-743658b2-1d2f-11e8-8af2-7a70debfc18b.png

— You are receiving this because you were mentioned. Reply to this email directly, view it on GitHub https://github.com/NGEET/fates/issues/339#issuecomment-369639758, or mute the thread https://github.com/notifications/unsubscribe-auth/AMWsQw9pDanW7nH-8PDtj3RjKwEJAKQlks5taBvxgaJpZM4SXP8C .

[serbinsh]

@rosiealice Hahaha! So that is all species within that dataset. And some are actually understory, but not many (the low end). Partly shown to highlight the data we have to inform params, but also we probably need to think about whether this first PFT is all spp or just mid to late successional evergreen and/or deciduous, etc. That fig lumps them all so not surprising its broad....using meta analysis we can narrow that down

[ckoven]

completely agree with @mdietze, which is basically what I meant to say. For the radiative diagnostics, do we have any sort of benchmark light profiles through the canopy that we can compare to? I remember we discussed this a while ago but forgot where we were with that.

[rosiealice]

Further to my leaf economics point, there is a good deal of uncertainty around the vertical gradient of leaf properties. Both LL and SLA have been shown to increase with depth. I'll try and find those (kitajima and Lloyd respectively) papers. If, however, NPP is actually negative for these lower canopy layers, then the leaf construction probably isn't what's most important. I keep thinking we need a way to look at the leaf layer level assimilation profile. This is presumably on the restart file as it accumulates for the trimming algorithm? I might try and make a script that can plot it out.

On Mar 1, 2018 9:31 AM, "Shawn P. Serbin" notifications@github.com wrote:

@rosiealice https://github.com/rosiealice Hahaha! So that is all species within that dataset. And some are actually understory, but not many (the low end). Partly shown to highlight the data we have to inform params, but also we probably need to think about whether this first PFT is all spp or just mid to late successional evergreen and/or deciduous, etc. That fig lumps them all so not surprising its broad....using meta analysis we can narrow that down

— You are receiving this because you were mentioned. Reply to this email directly, view it on GitHub https://github.com/NGEET/fates/issues/339#issuecomment-369649225, or mute the thread https://github.com/notifications/unsubscribe-auth/AMWsQ4uO34zPU2DZ1lZZGBhyeAe2AsjEks5taCJkgaJpZM4SXP8C .

[ckoven]

@rosiealice we even have these as non-default history variables already (from when I started down this path before I got distracted): TS_NET_UPTAKE_CNLF and YEAR_NET_UPTAKE_CNLF. also the various radiative terms PARSUN_Z_CNLF, PARSHA_Z_CNLF etc.

[huangmy]

All, I am adding a few comments from @mpaiao on my logging manuscript here, which are in general consistent with @ckoven's list but are more specific based on his experience with ED2. This may help us think through this:

Quote from @mpaiao:

1. I should probably know this, but how does FATES decide LAI of each cohort? Is this based on allometric equations relating size with leaf biomass, like in ED2? I’m asking this because I have been struggling with the leaf allometry in ED2, and I noticed that different allometric equations can easily change LAI by 2 m2/m2 or so.
2. Being very ignorant on the FATES implementation, I think this suggests that either too little light is reaching the understory, or that the light compensation point of understory trees is too high, or that the turnover rate of living tissues is too high.
3. Another possibility that I learned from the allometry drama in ED2 is that if you use equations based on trees with DBH>5cm and the minimum recruitment size in the model is much smaller than this, the allocation to pools can make seedlings to have high allocation to wood and low allocation to leaves, which makes the seedling lives really hard as they may have insufficient leaf area to get enough carbon to grow.
4. Another thing may be that canopy trees are growing boundless, so you have too dense canopy. This may be a symptom that other mechanisms should “prune” the canopy (like having branch turnover, higher canopy turnover rate, etc.).

[ckoven]

OK, wanted to update this thread in the context of radiation updates I made. I think radiation biases aren't the primary reason behind the inability to maintain LAI. A couple reasons for this: (1) as noted in https://github.com/NGEET/fates/pull/349#issuecomment-371671217 the radiation profiles look pretty close, and possibly brighter at depth, to the ones published by Mercado et al 2007 for Manaus. So, while it would be great if we had PAR profiles at BCI (does anyone?) a first cut suggests that radiation profiles aren't crazy and, if anything, brighter than they should be.

Using the new code and clumping index of 0.85, I tried doubling the allometric bleaf to see if the higher LAI would be maintained. Two relevant figures: (1) the profiles in the same form as shown above for the double bleaf (dashed line) and control (solid line), and (2) timeseries of LAI in the doubled bleaf (red curve) vs control (black curve) runs.

1:

2:

So the trimming is still kicking in and preventing high LAI, despite what seem to be reasonable light profiles. Note that those profiles were diagnosed in the first few years before the trim eliminated the deeper leaves.

So what all this says to me is that, while going into the radiation profiles was a necessary and useful exercise, it isn't actually the problem. And I don't think that the respiration throttling will (solely) solve it either, because that only applies to entire cohorts that would otherwise be in negative carbon balance; not leaf layers. So I think we may need to shift the focus of this problem to the leaf economics profiles, as per @rosiealice's suggestion yesterday. Right now, of the things that vary through the canopy, FATES has a sort of partial inclusion: it varies vcmax, jmax, and dark respiration, but it does not vary SLA or leaf lifespan. Thus the physiological operating costs and benefits of the leaves vary through the canopy, but not the construction costs.

A last figure, to this point: profiles of the net carbon balance of the leaves through the canopy, so corresponding to variable ts_net_uptake, which is effectively gpp minus dark respiration:

What the profile shows is that the leaves themselves remain (barely) in positive balance through the canopy down to an LAI of 4 or so. But the plant still isn't hanging on to them.

[ckoven]

Update to the net C uptake profile figure in prior post: the units as shown were wrong. I had thought there was a missing kg:g conversion, but looking closer I realize it was a second:dtime conversion, so those values on the figure are all too high by a factor of 1.8. Fixing the units for this in #349.

[ckoven]

All - I wanted to add another possible hypothesis to this thread about what might be going on in FATES that leads to the low leaf area bias, which is the assumption that FATES makes that leaf and fine root biomass are proportional to each other. What this assumption means in practice is that, when a given cohort is trying to decide whether to trim or de-trim its leaf area, it weighs the benefit of increased leaf area against the costs of both an increased leaf biomass and an increased fine root biomass. E.g., here: https://github.com/NGEET/fates/blob/master/biogeochem/EDPhysiologyMod.F90#L221

This calculation is really different from other ones that are exploring similar ideas in the literature, e.g. the Lloyd et al paper that we are using for our canopy leaf trait scaling profile (https://www.biogeosciences.net/7/1833/2010/), which weighs only the costs and benefits of leaf area on its own. So while it does make some intuitive sense that a given increment of leaf area might require a given amount of root area to support it, the evidence behind that is maybe not all that strong, and I think one could equally argue that a given root biomass ought to be required for a given water and nutrient uptake flux, both of which would decrease on the margin for a given increment of leaf area? Or, we could simplify further and just make root biomass follow a constant allometric relationship with DBH (possibly lower than our current value to reflect the root trimming that we've included thusfar), at least until we can actually calculate explicitly the benefits of an increment in root biomass in addition to its costs? In any case, I wonder if some of the behavior that we're seeing may be a consequence of this assumption, and so I'd propose that we include a test of what kind of LAI profiles the model would attain with fixed fine root biomass allometry.

[rgknox]

Another thing that we may want to keep in mind. Our dbh at maximum height is 150 cm right now, which is really describing an emergent (55 m) more than it describes a canopy tree (at least following our Obrien et al. height-diameter allometry). Since maximum height defines the capping point of maximum leaf biomass, I'm wondering what type of non-linear effects this has on the forests holding leaf holding capacity.

[rgknox]

Although, as Charlie pointed out in the fire call, the crown area exponent typically matches the biomass exponent, so the leaf density per m2 is linear as a tree grows?

[ckoven]

OK, as described in #354, I tried decoupling the trim logic in the leaf allometry from the fine root allometry. Here's a figure of the resulting timeseries of LAI: in both the black and red lines, I doubled the leaf:stem allometric coefficient; the black line is with the old allometry (fine root ~ trimmed leaf), the red line is with the revised allometry (fine root ~ untrimmed leaf). To be clear, this is just meant as a sensitivity test to identify if this change could solve the LAI bias, but the result is that it seems that it could.

[rosiealice]

SO red is (fine root ~ untrimmed leaf) AND the stem:leaf change? Just because I'd expect it to be more expensive, all else being equal, if one just swopped to tracking the maximum leaf (and thus root) biomass rather than the trimmed amount... Either way, this does look like carbon budgets and not necessarily light profiles might help with this problem.

2018-03-19 16:01 GMT-06:00 Charlie Koven notifications@github.com:

OK, as described in #354 https://github.com/NGEET/fates/pull/354, I tried decoupling the trim logic in the leaf allometry from the fine root allometry. Here's a figure of the resulting timeseries of LAI: in both the black and red lines, I doubled the leaf:stem allometric coefficient; the black line is with the old allometry (fine root ~ trimmed leaf), the red line is with the revised allometry (fine root ~ untrimmed leaf). To be clear, this is just meant as a sensitivity test to identify if this change could solve the LAI bias, but the result is that it seems that it could.

[image: screen shot 2018-03-19 at 2 52 10 pm] https://user-images.githubusercontent.com/10852790/37624643-10e542e0-2b86-11e8-98e0-8ae3a0509b99.png

— You are receiving this because you were mentioned. Reply to this email directly, view it on GitHub https://github.com/NGEET/fates/issues/339#issuecomment-374396714, or mute the thread https://github.com/notifications/unsubscribe-auth/AMWsQ1Ww4wd1tkL75pf2ZQx-ViasDau5ks5tgCqngaJpZM4SXP8C .

--

Dr Rosie A. Fisher

Staff Scientist Terrestrial Sciences Section Climate and Global Dynamics National Center for Atmospheric Research 1850 Table Mesa Drive Boulder, Colorado, 80305 USA. +1 303-497-1706

http://www.cgd.ucar.edu/staff/rfisher/

[ckoven]

@rosiealice right. in that test, I also reduced the fates_allom_l2fr term to try to conserve root biomass, though actually I overcorrected and actually reduced fine root biomass a bit relative to the control. But in this case, that shouldn't matter to the LAI profile per se because I started from inventory; if we had started from bare ground it would obviously have affected the growth rates. But by decoupling the two, we'll need to adjust fates_allom_l2fr so as to better match the GEM-type observations of fractional allocation to fine roots at the ecosystem level (which I suppose we'd need to do anyway).

[ckoven]

Wanted to update this thread with a couple relevant figures from the most recent branch that I've been working on, which includes #360, #354, and #349. Also I modified the parameter file from what I had been using earlier:

diff fates_params_baseline1_leafclumping_nomaxsize_newsai_2xbleaf_newbrootallom_withheight.c180308.cdl fates_params_baseline1_noleafclumping_nomaxsize.c180308.cdl
217c213
<       fates_allom_fmode:flag_values = 1.,2. ;
---
>       fates_allom_fmode:flag_values = 1. ;
261,262c257,258
<       fates_allom_sai_scaler:long_name = "allometric ratio of SAI to target bleaf" ;
<       fates_allom_sai_scaler:units = "m2/g" ;
---
>       fates_allom_sai_scaler:long_name = "allometric ratio of SAI to bdead" ;
>       fates_allom_sai_scaler:units = "m2/kg" ;
662c656
<  fates_clumping_index = 0.85, 0.85 ;
---
>  fates_clumping_index = 1., 1. ;
696c690
<  fates_allom_fmode = 2, 2 ;
---
>  fates_allom_fmode = 1, 1 ;
706c700
<  fates_allom_l2fr = 0.5, 0.5 ;
---
>  fates_allom_l2fr = 1, 1 ;
714c708
<  fates_allom_d2bl1 = 0.14, 0.14 ;
---
>  fates_allom_d2bl1 = 0.07, 0.07 ;
722c716
<  fates_allom_sai_scaler = 0.0012, 0.0012 ;
---
>  fates_allom_sai_scaler = 0.05, 0.05 ;

resulting profiles of PAR, and of net leaf C balance, through the deepened canopy and understory are:

---

---

Overall I think we are making progress towards deeper canopies. Turns out that #360 actually has a pretty big impact on where the understory net leaf C balance curve intersects zero; a run with partial implementation of #360, such that the understory was still using canopy scaling as if it were the canopy, looks like the following:

[ckoven]

this discussion is all out of date now. closing.