Describes the intermediate grid to which the transform is performed.


It may be coincident with the target grid, in which case there isn't a second interpolation and the transform to the target grid is direct. For other keywords controlling the interpolation see the diagram below.


Possible intgrid values are 


Disables any intermediate grid before interpolation, i.e. the transform will go directly to what has been specified with keyword GRID.

Disables any truncation before the transformation

In case a rotation is requested, INTGRID=NONE is ignored and defaults to INTGRID=AUTO.

AUTOAn intermediate FULL Gaussian grid is used, derived from the target GRID, by looking at the equivalent resolution. This is the default behaviour.


Specifies a Gaussian grid name like F80, O80, N80, N640 etc. (like [FNO][1-9][0-9]*), supported by the MARS client. Specifies the intermediate grid, e.g. O96 for an Octahedral Gaussian with 96 lines between Pole and Equator.

SOURCESets the intermediate grid to Octahedral Gaussian with N=(cubic relation from input truncation)=T+1, e.g. from input T1279 INTGRID=SOURCE(=O1280)
OFFReserved MARS value that resets the value inherited, effectively removing the keyword from the request which then results in intgrid being set to AUTO.


intgrid=F80      # Specifies a full Gaussian grid (with 80 lines between Pole and Equator)
                            # as intermediate grid.


This keyword cannot be specified. Behaviour is the same as the default for MARS, i.e. AUTO.


  1. Hi Carsten Maass ,
    do you have an idea whether the transformation spectral → grid is performed in single (float32) or double (float64) precision when I launch a MARS request?
    Can I also change this behaviour with the "accuracy" keyword?

    Many thanks

    1. These transformations are always done in double precision. The keyword accuracy doesn't affect the actual transformation. It only specifies the number of bits per value when written to the target GRIB message.