High Albedo bias in SP mode

[rosiealice]

From comparisons with iLAMB, it appears that FATES-SP has a high albedo and reflected shortwave bias.

We had originally thought that this could be tracked to the layering of the surface radiation scheme. Indeed, some fraction (a bit less than half) of this can be fixed by moving from a scheme with thick layers to one with thinner layers (see fig below where we move from the CLM5 comparison run (SP, with various modifications) to the 0.2 layer thickness FATES (big increase in reflected radiation, FSR), to variable thickness FATES with thin layers (not much change) to FATES with variable thick layers (a further large increase in reflected radiation).

but even with thin layers, there is a relatively large difference.
The good news is that the run time is so dominated by history variable machinations at this point that changing from the thin to the thick layering scheme barely changed the total cost ;)

There are three avenues to pursue:

1. making very very very thin layers and seeing if there is any configuration of layering where this goes away.
2. Looking at comparison runs of CLM5 and FATES-SP with NO STEM AREA. The error tracks a little with the ESAI, and there may be some critical differences in how the layering scheme comprehends SAI now compared to big leaf.
3. Look at whether these errors are within the parametric tolerance of the leaf and stem reflectance parameters and might be a tuning issue.

Here are the changes in SABV and SABG (veg and ground absorbed solar) for completeness

Noting that all of these figures are for the month of June, to speed up making these plots!

[rosiealice]

ILAMB comparisons here
https://webext.cgd.ucar.edu/I2000/FATES_SP_4x5test/lnd/_build_FATESSP_FATESnocomp/RadiationandEnergyCycle/Albedo/GEWEX.SRB/GEWEX.SRB.html

[rosiealice]

IMPACT OF SAI

Looking at explanation #2 illustrates something a little wierd, which is that the reflectance responds in the opposite direction to the removal of SAI in FATES and CLM5.

In both models, removing SAI reduces the amount of radiation absorbed by the canopy.

and increases the amount hitting the ground.

but in FATES removing stems reduces the reflected radiation. (So adding stems increases the reflectence. So the leaves+stems are 'shiner' than the leaves on their own).

This comparison of FATES with and without stems is with the thicker (1,0LAI) leaf layers. I will try it again with thinner layers to check it's not a feature of the grainy canopy.

[rosiealice]

HIGH RESOLUTION FATES CANOPY

Here we have

1. the CLM5 comparison.
2. FATES with 0.2 VAI thick layers
3. FATES with variable thin layers (top bin 0.025, inc 1.5)
4. FATES with variable fat layers (top bin 1.0, inc 2)
   

You can see that all the FATES runs have higher reflected radiation. The fat layer run is particularly high. The very thin layer sun does not get as low of a reflected flux as the CLM5 comparison run. It seems that the resolution of the layers may not be the source of the bias.

BUT the default simulation has very fat layers and a much bigger bias than it needs to. We should probably change the defaults to a version with a thinner top layer. Exactly what this is is likely another thread. ''

[rosiealice]

Impact of SAI on high res canopy.

Removing steam area in the context of a thin layer canopy gave similar results...

So adding stem area increases the light absorbed by the canopy, but doesn't impact the reflected fraction in FATES.

[rosiealice]

Impact of top layer bin width on radiation fluxes.

At the risk of this thread being too much to digest (!) here is a more systematic look at the impacts of @ckoven's new parameters on the fluxes.

In these simulations, we are changing the width of the top bin ,and holding the increase factor constant at 1.5.

There doesn't seem to be much to be gained from going <0.1, but 1. and 0.5 seem to introduce pretty large biases.

Also, this is really to confirm that the impact of resolution is saturating and won't solve the residual bias in albedo.

[rosiealice]

Impact of increase factor on fluxes.

Holding the width of the top bin at 0.1, this is the impact of changing the increase factor from 1.5 to 1.75 to 2.0. There is a small (order 1 watt) change from 1,5 to 1.75.

[ckoven]

@rosiealice what about 0.2 top-bin resolution?

[rosiealice]

You mean as the new default? With 1,5 as the increase factor? I think that should get rid of most of the layering induced biases...

[aswann]

My question might be way too naive, but are the NIR and VIS reflectances set the same for stems in both FATES and CLM? (I'm sure you thought of that already, but figured I'd ask anyway). I know @marysa found the NIR reflectances had a big influence on total albedo.

[rosiealice]

Indeed, I was very much hoping I had made a mistake in my attempts to make them uniform. Looking quite carefully, I think they really are the same.

[rosiealice]

@walkeranthonyp I was wondering whether your MAAT implementation of th Norman RTM might be an interesting way to baseline what should be happening here under some controlled condition?

[ckoven]

@rosiealice I was just suggesting a comparison of 0.2 top-level bin width alongside the ones that you've already done at 0.1, 0.5, and 1.0

[rosiealice]

OK right. Yes, can do that for sure.

[walkeranthonyp]

@rosiealice I can do that -- do you have particular conditions that you'd like to test? What is the current RT scheme in CLM? Also I don't have stem in the current MAAT RT scheme, so that would take longer to add.

A couple of questions:

• Is the LAI bin increase factor in units of LAI or is it some other kind of increase parameter?
• I got a little lost above. Is the latest that adding stem does not increase the reflected fraction in FATES anymore? It would make sense that it does increase reflected fraction as stem is more reflective (0.16 & 0.31) in the visible than leaves (0.1 & 0.7, different PFTs though). Stem is often less reflective in the nir (0.39/0.53 vs 0.45/0.35) but nir is a lower proportion of total incoming radiation.

[rosiealice]

Hi Anthony,
We should probably set up the details of this offline, but yes, it would be good to test the albedo predictions for a standrardized setup in terms of LAI, layering, solar angle etc.

There are two issues. The first is the bias in the albedo in general, and the second is the response to adding SAI. For the first, we can consider a 'leaf only' setup, since there is a mismatch with CLM5 for that as well as the condition with stems.

One thing is that, given the iterative solution, we might not be able to track down the nature of the biases without doing some relatively detailed interrogation of how, .e.g. the first pass of radiation through the profile is processed. Not sure how to manage that....

thanks for thinking about this :)
-Rosie

[rosiealice]

I guess this is now concluded, in the presence of @adrifoster's successful testing of the two stream code?

[adrifoster]

Not sure ... it's still high, just not as high

https://webext.cgd.ucar.edu/I2000/FATES_global_4x5/lnd/_build_FATES/RadiationandEnergyCycle/Albedo/CERES/CERES.html

[rosiealice]

This is resolved by the twostream radiation scheme.
#1036

but we perhaps need to make it default in the parameter file (and almost certainly should at this point) before we can close this out? @rgknox ?

[rgknox]

@rosiealice , while two-stream is stable, it still has some quirks that are giving restart testing issues. As far as I can tell, its subtle things like reproducing history output on the first time-step of the restart, etc, but we need to get this squared away before we set it to default.

[rosiealice]

OK fair enough... Does that mean we don't expect it to pass restart tests if we change to having it as the default?

[rgknox]

yes @rosiealice , I've been meaning to prioritize this so that we can change 2-stream to default. I've been focusing on things that affect calibration efforts answers, but this is something I think is important to get done soon so that we can protect that code better.