search

hollorol / RBBGCMuso

RBBGCMuso is a software package that supports the application of the Biome-BGCMuSo biogeochemical model.
GNU General Public License v2.0
20 stars 11 forks source link

paramSweep and musosensi #18

Closed science-explore closed 1 year ago

science-explore commented 2 years ago

I am a student who is very interested in ecology. I am very interested in your work. I am currently trying to test. I have an immature question, paramSweep and musosensi，are there any differences between these two functions？and why should they be developed separately? I understand that they are also used to obtain the parameters with the smallest error from the actual results. Otherwise, I can’t think of the difference.

zbarcza commented 2 years ago

Hello, I am glad that you are interested in the package. The answer is simple. paramSweep is a 'one-at-a-time' logic based sensitivity analysis. In fact this is a tool to learn some basic behavior of the model by adjusting one parameter only. The results is very informative. musosensi is a sophisticated and complete sensitivity analysis tool that implements a Monte-Carlo experiment and adjusts many parameters during all simulation steps. The results can be used to select parameters that are subject to Bayesian optimization. Note that none of the mentioned packages use observations. That is another story. Cheers, Zoltan

science-explore commented 2 years ago

think you, I understand

zbarcza commented 2 years ago

Hello,

For some reason I do not see your questions here. So I decided to quote them.

" 1. In my research, I am more concerned about the sensitivity of parameters to annual changes in gpp and npp. In the sensitivity test and parameter adjustment of your model, can you do this with simple operations?"

Yes. There is a way to do post-processing with your output variables. By default it uses average long term daily data, which will be proportional with the annual data. Please try it.

" 2. I am trying to optimize the parameters recently. I plan to use grid search for hyperparameter optimization. What do you think of this method? Can you give me some reference about GLUE's method?"

Grid search might be okay in case of low number of parameters and narrow intervals. It is global search, I guess. Not suitable for higher number of parameters. For such case Monte-Carlo method is more suitable.

GLUE has a large literature. Maybe try this one: https://onlinelibrary.wiley.com/doi/full/10.1002/hyp.10082

And this is my favorite: https://agupubs.onlinelibrary.wiley.com/doi/10.1029/2008WR006822

Cheers, Zoltan

zbarcza commented 2 years ago

Hello

Check this for answers

https://github.com/hollorol/RBBGCMuso/issues/18#issuecomment-945579022

Cheers, Zoltan

science-explore wrote:

Sorry to disturb you again, I have some other questions:

  1. In my research, I am more concerned about the sensitivity of parameters to annual changes in gpp and npp. In the sensitivity test and parameter adjustment of your model, can you do this with simple operations?
  2. I am trying to optimize the parameters recently. I plan to use grid search for hyperparameter optimization. What do you think of this method? Can you give me some reference about GLUE's method?

— You are receiving this because you commented. Reply to this email directly, view it on GitHub https://github.com/hollorol/RBBGCMuso/issues/18#issuecomment-945579022, or unsubscribe https://github.com/notifications/unsubscribe-auth/AEYL6ZSNZLTYDW4VRH2LJRDUHPR3VANCNFSM5FR7JYLQ. Triage notifications on the go with GitHub Mobile for iOS https://apps.apple.com/app/apple-store/id1477376905?ct=notification-email&mt=8&pt=524675 or Android https://play.google.com/store/apps/details?id=com.github.android&referrer=utm_campaign%3Dnotification-email%26utm_medium%3Demail%26utm_source%3Dgithub.