raphaeldussin/brushcutter

clean boundaries for MOM6 ocean model

brushcutter's Issues

Parent / Child relationships

1. Define Parent grid (e.g. global model)
2. Define Model Child grid (e.g. CCS)
3. Define OBC segments ( MOM6 specs )
4. Define "exchange" defined by the intersection of the interface positions on the child and overlapping parent cells.

*We need a grid nomenclature. Could be a (case-sensitive) copy of MOM6 grid type written in Python.

*Also, the Python parser needs to be able to read the MOM6 specifications referenced above. Otherwise, the user needs to specify the OBC's in two different places, and that is just wrong.

*The "exchange" grid format needs to be specified , and for clarity, the format should be the same in F90/Python.

reduce use of kwargs

mostly imin,imax,jmin,jmax in segment class

TRAVIS + Doc

simplifying obc_segment extent specification

Currently, obc_segment extent specification requires that the user explicitly enters the model grid indices. It would be helpful to avoid user error by allowing for symbolic specification of segment extents, as is done in the MOM6 segment parser, i.e. segments along the northern or western edges of the domain could be denoted by "N".

bulletproof reading 2d/3d array from netcdf

performance improvements

just a few thoughts on perf and what may happen next :

• target grid and locstream, regriding matrices are created for every segment
• regriding to locstream is only x2 faster than regriding to grid, did I miss something here ?
• currently it would be faster to regrid to full domain then extract segments rather than run a regriding for each segment but this breaks the logic of the implementation
• thinking about saving grid and regriding objects so that we don't have to generate them for every field.
• profiling necessary to find out where the code is spending too much time.
• is the extrapolation onto land fast enough ?

code factorization

duplicates between obc_variable and obc_vect_variable

read_field has no mechanism to do a local read

What is the best way to sample a NetCDF4 variable ? This needs to work for rectangles and lines

https://github.com/raphaeldussin/brushcutter/blob/master/brushcutter/lib_ioncdf.py#L132

improve localisation of corners for subset

• make sure a subset for arctic would work too

Special land values

The MOM6 boundary files were generated in such a way as to not have a land mask. This could well be intentional and the result of the flooding. However, the bottom depths seem to have odd values, such that the salinity is freshest at several bottom levels. The velocities also have a similar discontinuity. The source files have land down there, I expect. Is there any way to tell MOM6 that those values are not to be trusted?

Segment orientation

How do I know if the segment is on the left or right edge of the domain?

"imin,imax" as arguments for segment generation should be replaced with "istart,iend"

rotation of input velocities (if irregular)

if irregular grid, need to interp to T point, rotate

replace geometry attribute by has_vertical (or similar)