search

kdorheim / MENDplus

MEND R package and materials for the mechanistic reaction modeling analysis.
https://kdorheim.github.io/MENDplus/
0 stars 1 forks source link

Comparison data for optimization #6

Open kdorheim opened 4 years ago

kdorheim commented 4 years ago

Hi @jianqiuz could you add or direct me to some example comparison data we could use to optimize MEND?

jianqiuz commented 4 years ago

This is year long incubation from 3 NutNet sites. This files contains every you need to run MEND. We can run tests, but probably won't be able to put it on any pubs or packages as examples.

For more background information on the study, see attached papers SBB2016.pdf Riggs2015_Article_NitrogenAdditionChangesGrassla.pdf

nutnet_testing.xlsx

kdorheim commented 4 years ago

Thanks!

kdorheim commented 4 years ago

@jianqiuz Ah for optimization were you thinking about setting up a nonlinear optimization or some sort of Bayesian calibration?

kdorheim commented 4 years ago

Good question. I think in the long run, both of them are useful. An immediate use for a microbial-explicit model like MEND would be fitting experimental data (optimizing reaction kinetic parameters), in which case nonlinear optimization may or may not work. I have been fairly successful doing MCMC to infer kinetic parameters ( and initial conditions) under MATLAB. I think getting MCMC working would be the ultimate goal. For practical reasons, I think you may have a try nonlinear optimization, but if that takes more than one day and not working properly, we are probably better leaving that behind and do MCMC. Sometimes I found it either to constrain (optimize) MEND model with both respiration and biomass time series data. But the biomass data might not always available. For the dataset I have sent you, the initial conditions are generated from carbon size fractionation measurement, so the numbers are pretty reliable. You might work out an optimization routine that optimize the microbial uptake rate, and POM1&2 decomposition rates. The carbon use efficiency (related to Vd and mR) is an interesting parameter to play with as well. It renders strong control on projected CO2.

Hope this helps!