[MITgcm-support] Gradient check of adjoint sensitivities

[Nikolay Koldunov]

Dear all,

 I have a question concerning gradient check of adjoint sensitivities. I
have a small Arctic Ocean setup and calculate sensitivities of the last 30
day mean area and volume ( cost_ice_flag = 2 and 1) to air temperature and
wind forcing for one year.

 I run gradient check for one year for sensitivities to air temperature,
and results that I get from the finite difference and adjoint sensitivities
are reasonably (10-20% error) close to each other. However for
sensitivities to wind I get one-two order of magnitude difference. For
short time periods (up to 1 day) the difference is quite small, but
increase with time and become ridiculous (a factor of about 10 for volume
sensitivities to wind and a factor of about 100 for area sensitivities to
wind) for one year period. Also the sign is sometimes different.

Did anyone else experience this problem? Did you have any suggestion of
what could be the reason for such behavior?

 In the attachment results of grdchk (monthly mean volume vs vwind) for 1
year, 1 day, 10 days and 1 month. I will be happy to provide any additional
information you need.

 Thank you very much in advance,

Nikolay


-- 
Dr. Nikolay Koldunov
University of Hamburg
Institute of Oceanography
KlimaCampus

koldunovn at gmail.com
nikolay.koldunov at zmaw.de
ph. +49 40 42838 7580
Bundesstr. 53, room 143,
20146 Hamburg, Germany
-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://mitgcm.org/pipermail/mitgcm-support/attachments/20121211/55791917/attachment.htm>
-------------- next part --------------
A non-text attachment was scrubbed...
Name: grdchk_volume_vs_vwind.docx
Type: application/vnd.openxmlformats-officedocument.wordprocessingml.document
Size: 6243 bytes
Desc: not available
URL: <http://mitgcm.org/pipermail/mitgcm-support/attachments/20121211/55791917/attachment.docx>