All Aviso products are erroneously offset to the north and east by 1/2 grid cell

[jjr8]

Here is the email thread describing the problem:

From: AVISO [mailto:aviso@oceanobs.com] 
Sent: Tuesday, January 03, 2012 3:49 AM
To: jason.roberts@duke.edu
Subject: RE : [aviso] In the gridded data, to which point does lat/lon refer to? CRM:01213193

Dear Jason,

indeed, you are right: in the gridded data, the data are stored as 
points and not boxes that means that the lat/lon refers to the 
exact point of the given value (and not the SW corner as mentionned 
in the FAQ).

This is explained in the section 4.3 of the handbook named: 
"Ssalto/Duacs User Handbook: (M)SLA and (M)ADT Near-Real Time and 
Delayed Time Products" that you can find at  
http://www.aviso.oceanobs.com/en/data/tools/aviso-user-handbooks/index.html 

So, we will correct the FAQ asap.

Furthermore, the definition of the Mercator grid is defined in 
section 4.3.2 of the same handbook.  We apologize for not answering 
your question some years ago.

Thank you for releving the mistake and hoping this has answered 
your question

Best wishes to you and your family

Françoise Mertz
Aviso User Services

------------------- Message d'origine -------------------
De : ROBERTS, JASON
Date d'envoi : 12/12/2011
À : AVISO
Objet : [aviso] In the gridded data, to which point does lat/lon refer to?
Dear Aviso,
 
In the FAQ (http://www.aviso.oceanobs.com/en/services/faq/index.html), 
there is this question:
 
Question: In the gridded data, to which point lat/lon refer to?
Answer: The lon/lat given is the SW corner of the box, as specified 
in the metadata of the NetCDF file
 
Are you sure that answer is correct? Does the lat/lon refer to the 
center of the box instead?
 
I have been comparing geostrophic currents for recent years of 
Aviso DT-MSLA Upd to OSCAR (http://www.oscar.noaa.gov/). The two 
products produce similar estimates and have the same 1/3º 
resolution I have noticed that they appear to be "off" by 1/2 cell 
from each other. When I shift the Aviso image by 1/2 cell to the 
southwest, the alignment is much better. You can observe this 
effect by rapidly switching between the attached images. You will 
see that OSCAR.png and AVISO_Center.png correspond fairly well, 
while AVISO_SW_Corner.png appears to be "off" from both of those.

After observing this, I reminded myself that the Aviso FAQ was not 
necessarily wrong; it could be that I misaligned the OSCAR data. 
But after further investigation, I believe the Aviso FAQ is wrong 
and that the OSCAR data are correctly aligned. I have two reasons:
 
1.    The OSCAR coordinate system is simpler than Aviso, and 
therefore it is less likely that I misunderstand it. OSCAR uses an 
equirectangular projection (sometimes called plate carrée), with 
clear latitude boundaries of 80 ºN to 80 ºS. I am very confident I 
have understood their coordinate system and applied it correctly. 
Aviso uses a Mercator projection but does not anywhere specify the 
parameters of this projection for the merged satellite product (the 
FAQ does specify some reference ellipsoid parameters, but not for 
the merged product, and omits the other characteristics of the 
Mercator projection). As you may recall, I asked about the details 
of the Mercator projection two years ago and, unfortunately, I 
never received a satisfactory answer. (This is not your fault; 
perhaps if I spoke French I could communicate my confusion better.)
 
2.    If I apply the rule of the FAQ-that lat/lon refers to the SW 
corner of the box, rather than the center-to the Mediterranean 
data, is obviously poorly aligned with shoreline datasets such as 
GSHHS. It seems unlikely that the Med product would use coordinates 
of cell centers while the global product would use coordinates of 
the southwest corner.
 
In conclusion, I believe the FAQ is wrong. It should say that the 
lat/lon refer to the center of the box, not the southwest corner. 
Can you please confirm or deny this?
 
Thanks for your help,
 
Jason

All MGET tools that work with Aviso data were based on the erroneous answer in the Aviso FAQ. As a consequence, they improperly shift the data 1/2 cell to the north and east from the correct position. The solution is to remove the 1/2 cell shift.

[jjr8]

Fixed in r882, released in MGET 0.8a31.