Size Structured BCI Benchmarking

[rgknox]

This benchmarking exercise takes place at the BCI site. It make use of the local met driver data.

The benchmarks are generated via the scripts here (specifically, the branch called "rgknox-core-vars"): https://github.com/NGEET/tools_benchmarking_evaluation/tree/rgknox-core-ncvars

Working with @hmullerlandau, benchmark product is going through iterative quality improvements.

The comparisons use the acre toolset, found here: https://github.com/NGEET/acre/tree/census-bmarks

Also, see #407, for some small lessons learned about how to set-up a benchmark comparison (related to fusion, and canopy layering).

I encourage others to use these scripts, or the products of these scripts. For those who want the BCI benchmarks, two things: 1) We are still testing and working through quality control. 2 ) request the original dataset and read their guidelines

[rgknox]

Test 1:

In this first phase, I am running with dynamics turned on (non-st3), and leaf biophysics turned on as well (non-prescribed) and I am initializing with inventory. The expectation is that the simulation will drift from the census initial condition, and should reach equilibrium faster than starting at a true spin-up. (I have not tested if a bare-ground spin-up reaches a different equilibrium for similar parameters).

Here are the parameter files:

parameter file 1

parameter file 2

Summary on parameter files:

Based roughly on our tropical default set. Uses Chave above ground biomass and @isamcano 's height allometry, and default leaf biomass allometry. I turned off reproductive allocation until the plant is 20cm, upon which 0.1 of available carbon is allocated in set 1, and 0.05 of available fraction is allocated in parameter set 2. External seed rain is turned off.

The benchmarking script processes the last 10 years of the 20 year simulations. Benchmark report can be found found here:

bci-ptest-v12_plots.pdf

Tagging @adamhb @lmkueppers

My interpretation:

We sure are generating a lot of recruits... Although, the benchmark is for the 1cm class, and we need to generate a model diagnostic that is specifically recruits through that size class.

Clearly we are over-estimating mortality of small (very-much so) and large (less degree) size classes. It seems like the behavior of the modeled tree is more like that of an early successional, grows rapidly, has little shade tolerance.

Thoughts? Ideas? Things to try next?

[adamhb]

Hi Ryan,

FATES uses ~49 gC for the amount of carbon required to make a new recruit at 1.25 m in height (Charlie helped me find this number). I used FATES's allometric equations to find how much carbon would be in a 1 cm sapling in FATES: ~165 g. So, as you alluded to, the discrepancy between FATES's minimum size class and the BCI data could actually be responsible for a lot of the bias in recruitment rates shown in your PDF.

Your PDF shows ~339 recruits per ha per yr as the census benchmark. Does that include just canopy species? The settings xml file uses a default to inlude all species. I ran the benchmarking driver a few days ago for just the canopy species (dbh95 > 15 cm; 164 species) and got between 30 and 60 recruits per ha per year depending on the census interval (excluding the first interval). The "do_spp_rec" function has not changed in your latest updates to the benchmarking driver right?

Tagging @lmkueppers @rgknox

Adam

[rgknox]

Hi @adamhb ,

Thanks for the info on the size discrepancy. I need to look into adding a size structure recruitment benchmark to help clear this up.

The benchmarks I am showing are for all species. Although, I don't use the do_spp_rec function in my calculations. That could explain the differences in our estimates. I will double check my units.

[adamhb]

Hi @rgknox,

No worries. Just looked again and I used recruitment.eachspp which relies on the "recruitment" function. But I used only canopy trees which probably explains the discrepancy.

Adam

[rgknox]

Hi @adamhb, I'm currently working through a new series of tests. I changed a couple of parameters. In the DA King 1990 study at BCI, they studied saplings that were roughly 2.5 meters in height, which incidentally, using the Martinez-Cano height-diameter allometry equations, generates a plant that is just under 1cm in diameter. So I will be using 2.5 meters in height as my new minimum plant size. They also reported that that species ranged from around 100-200 grams of leaf biomass at this sampling size. I picked a canopy species from that study, Trichilia tuberculata, as a prototype. I think this will help craft a better starting size and allometry for non-seedling plants in FATES, and also generate more accurate comparisons with data. Will keep you updated.

Also next on my list is to add the recruitment rate diagnostic to the size structured output.

[adamhb]

Hi Ryan, That sounds great! Thanks so much for the update. Adam

On Mon, Aug 6, 2018 at 2:51 PM, Ryan Knox notifications@github.com wrote:

Hi @adamhb https://github.com/adamhb, I'm currently working through a new series of tests. I changed a couple of parameters. In the DA King 1990 study at BCI, they studied saplings that were roughly 2.5 meters in height, which incidentally, using the Martinez-Cano height-diameter allometry equations, generates a plant that is just under 1cm in diameter. So I will be using 2.5 meters in height as my new minimum plant size. They also reported that that species ranged from around 100-200 grams of leaf biomass at this sampling size. I picked a canopy species from that study, Trichilia tuberculata, as a prototype. I think this will help craft a better starting size and allometry for non-seedling plants in FATES, and also generate more accurate comparisons with data. Will keep you updated.

Also next on my list is to add the recruitment rate diagnostic to the size structured output.

— You are receiving this because you were mentioned. Reply to this email directly, view it on GitHub https://github.com/NGEET/fates/issues/409#issuecomment-410832012, or mute the thread https://github.com/notifications/unsubscribe-auth/AYN1oYofSLBOJEmYLaTzN_r9gq1KzMqMks5uOJ5GgaJpZM4VlnXt .

-- Adam Hanbury-Brown PhD Student Energy and Resources Group UC Berkeley