[Proj] Gauss conformal Lo33 on Clarke 1880 Cape Datum to WGS84 UTM36S - NTv2 file needed.

[Mikael Rittri]

Hello Zoltan,
You wrote:

> My question is:
> Did I specify the s_srs and t_srs parameters correctly

As far as I can see, yes (for what it's worth).

> Can I be certain that the SA.gsb grid was used?

I don't know if there is some tracing or debugging mode of ogr2ogr or gdalwarp that says
whether an NTv2 grid was used, since points outside the grid are silently non-datumshifted
(at least in Proj 4.8.0, I think).

However, for your point, I can verify that it was used:

With cs2cs, I get

> cs2cs +proj=tmerc +lon_0=33 +a=6378249.145 +b=6356514.966398753 +nadgrids=SA.gsb +to +init=epsg:32736

-49717.33 -2872272.46

450288.90 7128578.77 0.00

(when I have SA.gsb in the right directory), which is just a few centimeters from your results.
If I remove the +nadgrids parameter, or use the alaska.gsb instead, I get the result

450303.95 7128642.46 0.00

It is hard to know what's wrong, but maybe the SA.gsb grid shift file is not very accurate outside
South Africa, even though it extends outside the borders. The corrections outside South Africa
may have been computed from extrapolation rather than measurements.

(For example, I have noticed this for the conus grid shift file for NAD27: it has been extrapolated
to cover Cuba but is quite bad there.

So if I use +datum=NAD27, I get the default datum shift +nadgrids=@conus, at alaska, at ntv2_0.gsb, at ntv1_can.dat, and

> cs2cs +proj=longlat +datum=NAD27 +to +proj=longlat +datum=WGS84

-75 20

74d59'57.145"W 20d0'1.783"N 0.000

But if I use a datum shift that is more accurate on Cuba ("NAD27 to WGS 84 (88)", EPSG:15978), I get

> cs2cs +proj=longlat +ellps=clrk66 +towgs84=2.478,149.752,197.726,0.526,0.498,-0.501,0.685 +to +proj=longlat +datum=WGS84

-75 20

74d59'58.961"W 20d0'2.043"N -21.720

Horizontally, these differ by 53.4 meters.)

Anyway, back to Africa. If we don't trust the SA.gsb, we can try with the datum shift "Cape to WGS 84 (2)", EPSG:1129,
from the Directorate of Surveys and Land Information, Cape Town. "Residuals should not exceed 15 metres", says EPSG,
but I think they do.

> cs2cs +proj=tmerc +lon_0=33 +a=6378249.145 +b=6356514.966398753 +towgs84=-134.73,-110.92,-292.66 +to +init=epsg:32736

-49717.33 -2872272.46

450283.06 7128577.97 17.72

Or we can try with the datum shift "Cape to WGS 84 (1)", EPGS:1128, from U.S. Defense Mapping Agency TR8350.2,
September 1987, with accuracy 9 metres (confidence 67 percent, I think).

> cs2cs +proj=tmerc +lon_0=33 +a=6378249.145 +b=6356514.966398753 +towgs84=-136,-108,-292 +to +init=epsg:32736

-49717.33 -2872272.46

450286.20 7128578.80 17.88

Unfortunately, it seems that neither of these would make you very much happier.

As a last resort, if you have more examples of source and target points that are representative for your area of interest,
one could use the SEVENPAR program from Killetsoft to compute a local 3-parameter or 7-parameter transform.

Best regards,

Mikael Rittri
Carmenta
Sweden
http://www.carmenta.com

________________________________
From: proj-bounces at lists.maptools.org [mailto:proj-bounces at lists.maptools.org] On Behalf Of Zoltan Szecsei
Sent: Tuesday, August 28, 2012 10:46 AM
To: PROJ.4 and general Projections Discussions
Subject: Re: [Proj] Gauss conformal Lo33 on Clarke 1880 Cape Datum to WGS84 UTM36S - NTv2 file needed.

Hi All,
This issue goes on, slower than a snail.
[ --   Text deleted. See http://lists.maptools.org/pipermail/proj/2012-August/006407.html   -- ]

-------------- next part --------------
An HTML attachment was scrubbed...
URL: http://lists.maptools.org/pipermail/proj/attachments/20120828/4728eea5/attachment-0001.htm