Re: [ferret_users] inconsistent data grids: File needs axis attributes on coordinate variables indicating correct direction

[Ansley Manke <a.c.manke@xxxxxxxxx>]

    Hi Patrick, and others on this thread,
    
    Maybe there needs to be an FAQ about this. It's good to have this
    example.
    
    PyFerret creates grids for variables as it initializes a dataset. 
    It decides which dimensions have which direction, and as it works it
    requires that the directions are the same for all of the variables
    in the dataset. If there is any doubt about the directions of
    dimensions, it starts with the variables with the most dimensions,
    assigning the directions for their dimensions, and then works to
    create the grids for variables with fewer dimensions.
    If a dimension has a coordinate variable associated with it, the
    attributes are used to determine directions, and this is the most
    direct and preferred method. 
    
    If there is a coordinate variable, then direction is determined by
    looking at the coordinate variable's attributes. Once the direction
    is determined, the search stops, so for instance if we had a
    coordinate variable with an attribute :axis = "x" and :units="days"
    would be marked as an x axis.This is the order for checking
    direction information. 
    
    1- An axis attribute, with value x, y, z, t.  If there is
    no axis attribute, then
    
    2 - A cartesian_axis attribute.
    
    3- A cf_role attribute. This attribute is used in the
    discrete sampling geometries form for netCDF datasets.
    
    4- A _CoordinateAxisType attribute.
    
    5- If the direction is still undetermined, pyferret looks for a units
    attribute of the coordinate variable and checks whether those
    correspond to directions of longitude, latitude, depth/height (or
    pressure), or time.  
    
    6- A positive attribute.  positive=down or positive=up
    indicate a Z axis.

7- A standard_name attribute where the standard name value indicates longitude, latitude, height/depth, time.

    8- If the direction is still undetermined, or if the dimension does
    not correspond to a coordinate variable then a few obvious dimension
    names can indicate the direction:
    
         o Dimension names that are simple one-letter names X","Y", "Z",
    "T", "I", "J", "K", "L"
    
         o Dimension names that start with "XAX","YAX","ZAX","TAX"
    
         o Dimension names that start with
    "LON","LAT","TIME","DATE","DEP","ALT,"ELEV","HEI","LAYER","LEVEL"
    are assigned the appropriate direction. This last group is not
    applied if there is a units string that indicates something that is
    not a direction, in case there might be for example dimension that
    is say, "depletion", with units of "percent".

9- If the axis direction is still undermined, then they will be assigned, starting with the largest grid, working right to left with the first axis an x axis, then a y axis, and so on.  So the file here would look at bounds_lat(jmax, imax, nv)  and put nv in the x direction, imax in the y direction, and jmax in the z direction.  Then it cannot reconcile that with the 2-D grid (jmax, imax).

    In much older versions of Ferret, even more guessing was used to
    determine directions, so that any dimension name starting with T was
    put in the time direction and so forth. Probably jmax and imax would
    have been Y and X axes.  However this led to confusion and
    inconsistencies. For instance an axis called TSTATION was put in the
    T direction but should have been a simple axis in the X direction,
    for instance.

Ansley

On 9/29/2021 5:00 AM, Patrick Brockmann wrote:

Hi all,

I have investigated a little bit more on the internal ferret mechanism that leads to produce this error. 

**ERROR: inconsistent data grids: File needs axis attributes on coordinate variables indicating correct directions

Starting from an example given by the CF convention description from the folowing paragraph

https://cfconventions.org/Data/cf-conventions/cf-conventions-1.9/cf-conventions.html#cell-boundaries

I have reproduced the error from a very simple exemple.

$ cat notok.cdl

netcdf notok {
dimensions:
       jmax = 1 ;
       imax = 1 ;
       nv = 4 ;
variables:
       double lat(jmax, imax) ;
               lat:long_name = "latitude" ;
               lat:units = "degrees_north" ;
               lat:bounds = "bounds_lat" ;
       double lon(jmax, imax) ;
               lon:long_name = "longitude" ;
               lon:units = "degrees_east" ;
               lon:bounds = "bounds_lon" ;
       double bounds_lat(jmax, imax, nv) ;
       double bounds_lon(jmax, imax, nv) ;
       double sample(jmax, imax) ;
               sample:coordinates = "lon lat" ;
data:
lat = -28;
lon = 157;
bounds_lat = -29, -29, -27, -27;
bounds_lon = 156, 158, 158, 156;
sample = 1 ;
}
$ ncgen -o notok.nc notok.cdl

$ ferret

yes? use notok.nc

          *** NOTE:
          *** NOTE: Could not adjust grid for variable lat
**ERROR: inconsistent data grids: File needs axis attributes on coordinate variables indicating correct directions

The same file with modified dimension names (jmax, imax replaced by y, x) can be openned.

$ cat ok.cdl

netcdf ok {
dimensions:
       y = 1 ;
       x = 1 ;
       nv = 4 ;
variables:
       double lat(y, x) ;
               lat:long_name = "latitude" ;
               lat:units = "degrees_north" ;
               lat:bounds = "bounds_lat" ;
       double lon(y, x) ;
               lon:long_name = "longitude" ;
               lon:units = "degrees_east" ;
               lon:bounds = "bounds_lon" ;
       double bounds_lat(y, x, nv) ;
       double bounds_lon(y, x, nv) ;
       double sample(y, x) ;
               sample:coordinates = "lon lat" ;
data:
lat = -28;
lon = 157;
bounds_lat = -29, -29, -27, -27;
bounds_lon = 156, 158, 158, 156;
sample = 1 ;
}
$ ncgen -o ok.nc ok.cdl

$ ferret

yes? use ok.nc

I would like to understand why ?

Best regards

Patrick

--
Data Analytics and Visualization Engineer / Project Manager
LSCE/IPSL, CEA-CNRS-UVSQ laboratory
LSCE - Climate and Environment Sciences Laboratory
IPSL - Institut Pierre Simon Laplace
--

---

De: "Patrick Brockmann" <patrick.brockmann@xxxxxxxxxxxx>
À: "ferret users" <ferret_users@xxxxxxxx>
Cc: "Thanh-Lam Nguyen" <sebastien.nguyen@xxxxxxxxxxxx>
Envoyé: Lundi 27 Septembre 2021 16:48:27
Objet: [ferret_users] inconsistent data grids: File needs axis attributes on coordinate variables indicating correct direction

Hi ferreters,

A colleague has encountered a strange problem as he was trying to

nicely respect the netCDF CF conventions to describe a curvilinear grid (NEMO configuration).

ferret complains about some inconsistent data grid and missing axis attributes.

use notok.nc 
          *** NOTE: 
          *** NOTE: Could not adjust grid for variable mask_T 
**ERROR: inconsistent data grids: File needs axis attributes on coordinate variables indicating correct directions

Stranger, if I change the name of the dimension, then the file

is read. I have not been able to fix this by adding a direction attribute as requested.

The file is simple as:

$ ncdump notok.nc
netcdf notok { 
dimensions: 
       y_grid_T = 1 ; 
       x_grid_T = 1 ; 
       nvertex_grid_T = 4 ; 
variables: 
       double bounds_lat_grid_T(y_grid_T, x_grid_T, nvertex_grid_T) ; 
       double bounds_lon_grid_T(y_grid_T, x_grid_T, nvertex_grid_T) ; 
       byte mask_T(y_grid_T, x_grid_T) ; 
               mask_T:cell_measures = "area: area_grid_T" ; 
               mask_T:coordinates = "nav_lat_grid_T nav_lon_grid_T" ; 
       double nav_lat_grid_T(y_grid_T, x_grid_T) ; 
               nav_lat_grid_T:standard_name = "latitude" ; 
               nav_lat_grid_T:long_name = "Latitude" ; 
               nav_lat_grid_T:units = "degrees_north" ; 
               nav_lat_grid_T:bounds = "bounds_lat_grid_T" ; 
       double nav_lon_grid_T(y_grid_T, x_grid_T) ; 
               nav_lon_grid_T:standard_name = "longitude" ; 
               nav_lon_grid_T:long_name = "Longitude" ; 
               nav_lon_grid_T:units = "degrees_east" ; 
               nav_lon_grid_T:bounds = "bounds_lon_grid_T" ; 
data: 
bounds_lat_grid_T = 
 -29.5895932339965, -29.5895932339965, -27.835621888657, -27.835621888657 ; 
bounds_lon_grid_T = 
 156.999148343641, 158.999201071309, 158.999201071309, 156.999148343641 ; 
mask_T = 
 1 ; 
nav_lat_grid_T = 
 -28.7162844516478 ; 
nav_lon_grid_T = 
 157.999608915878 ; 
}

If I remove _grid_T suffix then the file is read correctly.
$ ncgen -o notok.nc notok.cdl
$ sed -e 's/_grid_T//g' notok.cdl > ok.cdl
$ ncgen -o ok.nc ok.cdl


Any idea on why there is the unexpected behaviour (test on names of dimensions) ?

Patrick

--
Data Analytics and Visualization Engineer / Project Manager
LSCE/IPSL, CEA-CNRS-UVSQ laboratory
LSCE - Climate and Environment Sciences Laboratory
IPSL - Institut Pierre Simon Laplace
--sed -e 's/_grid_T//g' tyty.cdl > tyty1.cdl