Backwards trajectory

Discussion about tangent linear and adjoint models, variational data assimilation, and other related issues.

Moderators: arango, robertson

Post Reply
Message
Author
User avatar
susonic
Posts: 160
Joined: Tue Aug 21, 2007 5:44 pm
Location: Jeju National University
Contact:

Backwards trajectory

#1 Post by susonic » Thu Mar 10, 2011 1:38 am

Dear ROMS users,

I was wondering whether ROMS can track back the orginal source place using adjoint model with floats?

-JH

User avatar
jivica
Posts: 136
Joined: Mon May 05, 2003 2:41 pm
Location: The University of Western Australia, Perth, Australia

Re: Backwards trajectory

#2 Post by jivica » Fri Mar 11, 2011 12:45 am

Think UCLA group has offline float tracking code for ROMS that is capable of going back in time.
So, basically you feed it with ROMS precomputed history files and use similar input file for initial float positions. There is a option for going frwd or back in time..
Capet and Sasa did that long time ago, I do not know if they have "new version" because that old one was using cppdefs.h and so on...
In that way you do not need ADJOINT
Hope it helps

rduran
Posts: 139
Joined: Fri Jan 08, 2010 7:22 pm
Location: Theiss Research

Re: Backwards trajectory

#3 Post by rduran » Fri Mar 11, 2011 2:04 am


User avatar
susonic
Posts: 160
Joined: Tue Aug 21, 2007 5:44 pm
Location: Jeju National University
Contact:

Re: Backwards trajectory

#4 Post by susonic » Fri Mar 11, 2011 3:45 am

Thanks for your notes, Ivica and Rodrigo.

So, I've got two options.

1. Modify the step_floats.F and interp_floats.F. and run the ADJOINT model with using floats.
2. Use history file to trace back.

drivas
Posts: 29
Joined: Tue Aug 28, 2007 8:31 pm
Location: CICESE

Re: Backwards trajectory

#5 Post by drivas » Fri Mar 11, 2011 11:38 am

That's right. If you decide to use option 2, you should save your model outputs as often as possible, so that the integration step is small enough to get not-so-large errors. But you should work with multi-trajectory ensembles anyway. For some details, you can see:

http://journals.ametsoc.org/doi/abs/10. ... 0JPO4327.1

In that paper all the results are reported in a statistical basis. A river discharge is also included.
Regards,
drivas

User avatar
susonic
Posts: 160
Joined: Tue Aug 21, 2007 5:44 pm
Location: Jeju National University
Contact:

Re: Backwards trajectory

#6 Post by susonic » Fri Mar 11, 2011 1:05 pm

Thanks for your message and link, David.

I'll follow your suggestion.

Regads,

-JH

User avatar
aneeshcs
Posts: 12
Joined: Tue May 31, 2005 4:17 pm
Location: Scripps Institution of Oceanography
Contact:

Re: Backwards trajectory

#7 Post by aneeshcs » Sat Mar 12, 2011 1:02 am

There is also one more option, which is to use adjoint passive tracers such as done in

Song, H., A. J. Miller, B. D. Cornuelle and E. Di Lorenzo, 2010: Changes in upwelling and its water sources in the California Current System driven by different wind forcing. Journal of Physical Oceanography, accepted.

and

Combes V., F. Chenillat, E. Di Lorenzo, P. Riviere, S. Bograd and M. Ohman, 2010: Cross-shore transport varaibility in the California Current: Ekman upwelling vs. eddy regime. Journal of Physical Oceanography, submitted. ( you can get the PDF from Manu's webpage, http://ocean3d.org/manu/pubs.html)

and

Chhak, K. and E. Di Lorenzo, 2007: Decadal variations in the California Current upwelling cells. Geophys. Res. Lett., 34, L14604, doi:10.1029/2007GL030203.

The ROMS Code for passive tracers is adjointed already. So you have to give a forcing file for your adjoint file with injecting passive tracers in the place you are interested in and the adjoint solution will give you the sensitivity of the regions which influenced your source of tracers.

drivas
Posts: 29
Joined: Tue Aug 28, 2007 8:31 pm
Location: CICESE

Re: Backwards trajectory

#8 Post by drivas » Sat Mar 12, 2011 5:19 am

If the Adjoint model is completely implemented, and if you can run your model whenever you want, it's most probably better to use the Adjoint model. I can tell you, however, that comparisons between the results shown in Rivas and Samelson (2011) and passive tracers advected forward in time internally in the model (done by rduran) show that both results are actually consistent. So, in summary, if you can use the Adjoint method, go ahead, but the offline method can be also a good option.
Good luck,
drivas

Post Reply