[Proj] GeographicLib geoid calculations

Charles Karney charles.karney at sri.com
Wed Oct 17 04:49:57 EST 2012

On 10/17/2012 04:22 AM, support.mn at elisanet.fi wrote:
> Hello,
> how accurate is the GeographicLib geoid separation calculation
> anywhere on the earth?
> Is it better than the EGM96 fortran 15 arc min bilinear implementation
> which requires 2 Mb memory for the table?
> Regards: Janne.

The errors for various geoid models and various grid sizes are listed in


For example for the 15' EGM96 grid using cubic interpolation the max and
RMS errors are 169mm and 7mm.  However, I would normally recommend the
5' EGM95 grid or the 2.5' or 1' EGM2008 grid.

GeographicLib uses 2 bytes per grid point to store the data (vs 4 bytes
per grid point for most other implementation) and, by default, does not
read the entire table into memory.

(Incidentally, GeographlicLib lets you work directly with the spherical
harmonic expansions of the geoid models using the Gravity class.  So you
can construct grids for any of EGM84, EGM96, or EGM2008 at any

More information about the Proj mailing list