possible bug in stations output: multiple conversion lon/lat -> fractional grid coords

[stef]

Hi, this is a follow-up to

viewtopic.php?t=6643https://www.myroms. ... php?t=6643

I just noticed that ad_initial.F calls grid_coords.F, which I believe converts
lon/lat to fractional grid coords. However, this has already been done
in the preceeding nonlinear forward model initialization, leading to
incorrect station locations.

I'm using RBL4DVAR RPCG.

I noticed the FLOATS_NOT_YET preprocessing flag in ad_initial.F, does
this mean "convert the float coordinates to fractional grid coordintates,
if it has not been done before"? I have not yet understood the floats.
If so, should there be a similar logic for stations?

I'm still assuming that the strategy is to append station data to for each
nonlinear run to the already existing station data file, and that the user
is responsible for splitting the chunks themselves. Is this actually right?

[stef]

Meanwhile I have figured out that the *_NOT_YET are hidden-feature flags and
FLOATS_NOT_YET does not mean what I thought.

I think I fixed my problem temporarily for my particular setup by

* calling grid_coords() only in the first of the two nonlinear run's initial.F
* uncommenting the call to grid_coords() in ad_initial.F and tl_initial.F

I should probably do it differently so I can still use the drifters/floats, but I'm not using them at the moment.

[arango]

The NOT_YET suffix in the CPP option means that such an algorithm is only available/executed in the nonlinear (NLM) kernel in applications that also activate the tangent linear (TLM) and adjoint (ADM) kernels. In such cases, we don't have the respective TLM and ADM transformations.