Integrate calibration pipeline into ClimaLand #621
Conversation
AlexisRenchon
force-pushed
the
ar/calibrate
branch
from
2a4b8d3
to
197737c
Compare
AlexisRenchon
force-pushed
the
ar/calibrate
branch
from
509a136
to
5e733a7
Compare
AlexisRenchon
force-pushed
the
ar/calibrate
branch
from
4579aab
to
c2add9d
Compare
This commit adds a simple calibration example, with detailed explanations. We calibrate two parameters, g1 and g0, on canopy transpiration data target, using EnsembleKalmanProcesses.jl. We use a single site model, at the ozark site, and calibrate on simulation output, with the goal of demonstrating the ability to recover the original parameters and demonstrate a simple calibration framework.
AlexisRenchon
force-pushed
the
ar/calibrate
branch
from
f395e2b
to
423c5b7
Compare
AlexisRenchon
requested review from
odunbar,
nefrathenrici,
kmdeck and
Sbozzolo
| # Packages | ||
| import ClimaLandSimulations.Fluxnet as CLS # to run the model | ||
| import EnsembleKalmanProcesses as EKP # to perform the calibration | ||
| import Random # we need it for ... |
| params = CLS.ozark_default_params(; conductance = CLS.conductance_ozark(; g1 = g1, g0 = g0)), | ||
| )[1]; | ||
|
|
||
| inputs = CLS.make_inputs_df("US-MOz")[1]; |
There was a problem hiding this comment.
What are these inputs and outputs? Can we access transpiration directly from sv?
|
|
||
| simulation_output = CLS.make_output_df("US-MOz", sv, inputs); | ||
|
|
||
| transpiration = simulation_output.transpiration .* 1e9 # 1e9 because very small otherwise |
There was a problem hiding this comment.
Why the 1e9? Is there an issue with floating point precision, or is there an error in the output?
There was a problem hiding this comment.
This also changes the float type
|
|
||
| # parameters prior | ||
| prior_g1 = EKP.constrained_gaussian("g1", 221, 100, 0, Inf) # mean of 221 sqrt(Pa) = 7 sqrt(kPa), std of 100 (3 kPa) | ||
| # Ollie: why does it return μ=5.3, σ=0.4? Are the values transformed? Is this to be of same order of magnitude? |
| # generate the initial ensemble and set up the ensemble Kalman inversion | ||
| N_ensemble = 5 | ||
| N_iterations = 5 | ||
| rng_seed = 41 # Ollie: why 41? |
There was a problem hiding this comment.
can erase? I dont think this matters too much
There was a problem hiding this comment.
In general - maybe should remove "Ollie" questions for now?
There was a problem hiding this comment.
Sorry I should've left a comment - this PR is not ready to be merged as I want to discuss these fews questions with Ollie first, but I think he is away at the moment.
But I still thought it would be valuable to review at this stage
| theta_true = [141.0, 0.0001] | ||
| fig = Figure() | ||
| ax = Axis(fig[1, 1]) | ||
| trange = 1:1:length(transpiration_target) |
There was a problem hiding this comment.
could note that this is not a physical time? "trange"
| @@ -0,0 +1,96 @@ | |||
| # This script describes the simplest possible calibration of ClimaLand, | |||
There was a problem hiding this comment.
Should we add this to buildkite? and upload plots
There was a problem hiding this comment.
or this could be a tutorial?
|
|
||
|
|
||
| # Plot | ||
| using GLMakie |
There was a problem hiding this comment.
Can we use CairoMakie? It is the more robust backend
This PR adds calibration experiments, starting with a minimal working example.