Hello Paul,
below I have copied comments from Vincent (any errors due to translation
are mine) and Mathieu to your remarks.
As mentioned before, I am not an expert and am just trying to relay the discussion from the French mailing list to the developers in order to find the best solution.
Paul Kelly wrote:
Hello Moritz
On Tue, 22 Nov 2005, Moritz Lennert
- Should I change the two files to reflect the correct defintions. Mathieu Basille proposes the following changes:
In datum.table : # Nouvelle Triangulation Française ntf "Nouvelle
Triangulation Française" clark80IGN dx=-168.0 dy=-60.0
dz=320.0
First of all that should be "Nouvelle_Triangulation_Francaise" even if it is not quite grammatically correct. It should correspond to the
official EPSG name which only uses ASCII characters. I have now added a comment to datum.table to indicate that.
Ok, thanks for changing that (although I didn't find the comment you
said you added).
In ellipse.table : # Clarke 1880 IGN clark80IGN "Clarke 1880 IGN" a=6378249.200 f=1/293.466021294
Two things: 1) The differences are so small from the version already in GRASS that it's not clear to me how big a difference it would make
(although I could be wrong). What is the problem that resulted in
the proposal for these changes?
Mathieu:
I don't know exactly the induced problems (probably a couple of meters...). But I really think that the problem is not there. The Clark80 as defined in GRASS is the same that the one used by ArcView (if I'm not wrong) and therefore is the most used in France. The problem is that it doesn't correspond to the OFFICIAL parameters that
we SHOULD use in France, to do things right. It's more a problem of principles than a problem of real troubles in the projections. If GRASS is supposed to do a good work, it has to implement the correct and official parameters.
2) I would rather fix the existing clark80 definition to get it right
than introduce an alternative different version.
Mathieu:
I agree on that point. I just added a second one to be able to import
ArcView layers (with the Clark80 ellipsoid) in different ways. But I'm not sure that's still needed with the option "Create a new location" when you import a layer.
So the question is whether the current definition of the Clark80 ellipsoid is useful for anything else. In datum.table, the carthage datum also references it.
On http://www.globalsecurity.org/military/library/policy/army/fm/6-2/appe2.htm (random google search) at the bottom, note 1, reads:
"a. Note 1. Any entry reading SEE NOTE ONE in Tables E-3 through E-25 of this appendix are so noted because of inconsistent listings of datums referenced to the Clarke 1880 ellipsoid. Table E-1 lists five different Clarke 1880 ellipsoids. DMA has adopted only one. Different countries have adopted different dimensions for the Clarke 1880 ellipsoid. These differences depend on two things: which of Clarke's original numbers were used ([a, b] or [a, f]) or which foot-to-meter conversion was used.
(1) In areas referenced to the ARC 1950 datum, the Clarke 1880 dimensions adopted are shown below.
a: 6378249.145326 b: 6356514.966721 f: 1/293.4663076
(2) In areas referenced to Carthage, Merchich, and Voirol datums, the adopted dimensions are shown below.
a: 6378249.2 b: 6356515.0 f: 1/293.46598
(3) The DMA-adopted dimensions are shown below.
a: 6378249.145 b: 6356514.8696 f: 1/293.465
(4) DMA TM 8350.2 with Insert 1 lists datum transformation parameters for local datums referenced to the DMA-adopted Clarke 1880 and not the dimensions adopted by other countries. Any datum with SEE NOTE ONE in the DDCT CODE column should be transformed to other datums with the user-defined option. "
So there might no be one single "correct" definition of this ellipsoid.
For better datum transformation you can also use nadgrids: france.gdb
in the PROJ_INFO file (instead of dx,dy,dz) Get the france.gsb file at http://www.stjohnspoint.co.uk/gis/france.htm I should maybe see about integrating this file into the GRASS distribution as it is hard
to find good datum transformation parameters for France.
Vincent:
Great to see that others have worked on the problem of creating a conversion grid NTF->GRS80 !
One other thing to look out for is the way to specify the Lambert Conformal Conic projection with GRASS/PROJ. I suspect g.setproj has some bugs with regard to this as there is more than one way of specifying the parameters for this projection. But I'm not too sure about it all, just have an uneasy feeling!
Vincent:
The Lambert Conformal Conic projections used in France are: - either
tangent with scale factor on both sides of the automecoic parallels -
or secant (equivalent definitions) You can find more information on
the website of the IGN,
See for example http://www.ign.fr/telechargement/FAQ/FAQ10.pdf (in english) and http://www.ign.fr/telechargement/FAQ/FAQ5.pdf.
- Or is this handled by proj ? What is the relation between the two
?
The ellipse.table file in GRASS dates from the mid-late 1980s and was
originally only used by some d.* commands that converted UTM co-ordinates to lat/long. It pre-dates the introduction of PROJ into GRASS in the early 1990s as far as I know.
Mathieu:
I have to admit that I'm totally lost at this point. I don't understand at all the relation between GRASS and PROJ... I thought that setting the good ellipsoid and datum in GRASS and then setting precisely the LCC projection used in France (with the right parameters) would suffice.
I'll try to play a bit with the france.gdb file... But I'm not sure to understand exactly how it can help me. Do I really need to use the
RGF93 datum for french projections ??? Uh... It's definitely not as simple as I expected... What a mess to use french projections for my job !
Moritz