[GRASS-user] "reboxing," or 3D regridding

[Tom Roche]

summary: I'd appreciate advice regarding tools and methods for
transforming data attributed to voxels in an unprojected global grid
onto a projected 3D grid with different horizontal and vertical
resolution (or pointers to other resources to consult).

details:

ESMF defines well (if somewhat oddly) the general problem:

http://www.earthsystemmodeling.org/esmf_releases/public/ESMF_5_2_0rp1/ESMF_refdoc/node3.html#SECTION03020000000000000000
> Regridding, also called remapping or interpolation [or resampling], is
> the process of changing the grid that underlies data values while
> preserving qualities of the original data.

ESMF seems to provide excellent tools for doing 2D regridding (or
interpolating data values from the cells/pixels of one 2D/horizontal
spatial grid to another), as does GRASS::r.proj

http://grass.osgeo.org/grass64/manuals/html64_user/r.proj.html

though I have not used either, and am quite new to GRASS. (My current
personal favorite regridding tool is the R package 'raster': see code @

https://github.com/TomRoche/GEIA_to_netCDF/

) However I'm not seeing tools for "reboxing," or interpolating data
values from the boxes/voxels of one 3D/horizontal+vertical spatial grid
to another. Am I missing something? I _do_ see (thanks, Doug Newcomb)
raster3D

http://grass.osgeo.org/grass64/manuals/html64_user/raster3D.html

but I don't see r3 API that does what I want:

I have output from a global atmospheric model that I'd like to use as
initial/boundary conditions for a regional model. This unprojected
"global input" (from the perspective of this usecase) netCDF has
dimensions=2.5° lon x 1.875° lat x 56 vertical levels. The regional
model covers North America using a 12-km grid projected LCC (Lambert
Conic Comformal), with 34 vertical levels: details @

https://github.com/TomRoche/cornbeltN2O/wiki/AQMEII-North-American-domain#wiki-EPA

The top height of the "regional output" is less than that of the global
input; i.e., the input domain fully contains the output domain, in all
3 dimensions.

Each box/voxel of the global input grid contains an estimate of its N2O
concentration. From those data I want to compute the concentrations for
each output box. I'd appreciate your recommendations for tools that can
do this. The best tool I've seen so far is R package=gstat, but (IIUC)

- gstat expects projected input. I'm not sure if I can work around that
  for this usecase. Is there a conservative projection over North
  America to which I could safely transform values from lon-lat
  (essentially via cropping?) in order to input them to gstat?

- as the name implies, 'gstat' is doing geostatistical (variogram- and
  covariance-based) modeling. I'm not sure either how to setup the
  distance weighting for my usecase. I'm also unconvinced that a
  statistical approach is necessary for this usecase, though it may be a
  sufficient, or the best-available, approach; furthermore my position
  may just be a prejudice due to my statistical ignorance.

TIA, Tom Roche <Tom_Roche at pobox.com>