[Thread Prev][Thread Next][Index]

Re: How to compress NetCDF float variables to 2-byte integer usingFerret 5.80?



Hi Gus,
I'll take some time for you to get a feel for how Ferret works, but you
sound like you're well on your way.  The manual is a large document, and
it can be hard to find what you need. The Users List is a good resource,
and searching the archives can sometimes turn up useful answers too.

We've had requests for the ability to work with other data types - there
are those of course who would like to do their analysis in double
precision, which would be a larger change than an integer output.
Perhaps writing 2-byte or other data types is a capability could be
added as a Ferret external function.

Ansley

Gus Correa wrote:

Hello Ansley Manke and Ferret users list

Thank you, Ansley, for your help.
Your clarification on the concept and use of the "immediate mode" of a variable (the value stored on a variable),
by contrast to the variable itself (a reference or pointer to a structure perhaps?),
was very insightful, and helped me understand how Ferret works.

It looks like that I can't use Ferret for the 2-byte integer compression of NetCDF files, as I intended to do.
I'll have to get around with some scripting using NCO tools,
or the hard way, writing a Fortran or C program using the NetCDF libraries directly.

I was trying to minimize the number of programs I need to use to post-process climate model output.
Since I need to use Ferret to do more complicated things (e.g. convert from sigma level to pressure coordinates),
I thought it might be possible to use it to do the compression as well.
Ferret has all the right ingredients for this, particularly the easy way to find the global max and min of a dataset.
The only missing feature is the possibility to output NetCDF data in different numeric formats.

Would it be appropriate to request this feature for a future release of Ferret?
(I.e, flexible numeric output format of NetCDF files, allowing floats, double, short, etc).
Or is it far from the goals of Ferret?

Thanks again.
Gus Correa

Ansley Manke wrote:

Hi Gus,
Welcome to Ferret. (just a note that when you send questions about using Ferret they should go just to ferret_users; contact_ferret is more for system-related things, installation questions, or reporting crashes or bugs and so on). You've written a great question, showing what you're trying to do, what happens when you try and what you need to know. It makes it much easier to answer!

First off, Ferret writes data to netCDF files only in the float datatype, so I'm not sure this whole exercise will save you much on the size of the output file.

The apparent discrepancy you see in the documentation between SET VAR/SCALE and SET VAR/SCALEFAC is due to the fact that Ferret commands and qualifiers can always be abbreviated to 4 characters (or fewer), so these two specifications are the same. See "commands, command syntax" in the Users Guide index for a list of notes about commands syntax.

SET VAR/SCALE and SET VAR/OFFSET are only for changing the scale and/or offset values which will be used when Ferret reads data from a netCDF file. Ferret does not write the scale factor or add_offset attributes when it writes a netCDF File; it writes the variable after applying all scaling. The RETURN keywords nc_offset and nc_scale are ways to access the values of these attributes as they appear in a file that you are reading, and RETURN keywords user_offset and user_scale are ways to get the value of any settings the user has made. These keywords are not things you can set, but just names for things you can get back from Ferret.

The syntax errors you're getting are because you want to use the variable my_scale as a scalar in the SET VAR/SCALE command. When the command is being parsed, Ferret does not generally evaluate the value of expressions. To ensure that the value is viewed as a scalar when the command is being parsed, enclose it in grave accent marks which forces the expression my_scale to be evaluated. This point is called "immediate mode" expressions. (We can add an example to the Users Guide to show this).

yes? let my_scale = intrange/(my_max - my_min)
yes? set var/scale=`my_scale` my_new_var


-Ansley

Gus Correa wrote:

Hello Ferret support person and users

I am a new Ferret user.
I would like to use Ferret 5.80 to achieve a (modest) compression of NetCDF files,
by doing this pretty standard procedure:

(1) Read the floating point or double precision variables from NetCDF files;
(2) scale them (i.e. apply the appropriate scale and offset factors) to fit the range of two-byte integers;
(3) convert them to two-byte integer variables, and
(4) write them to (hopefully smaller sized) NetCDF files.

Step (1) is just the Ferret command "use".
Step (2) goes more or less like this:

let my_max=my_var{x=@max,y=@max,z=@max,t=@max]
let my_min=my_var{x=@min,y=@min,z=@min,t=@min]
let intrange=65534.0
let new_missing_value=-32768
let my_scale=intrange/(my_max - my_min)
let my_off=(my_max + my_min) / 2.0
let my_new_var = my_scale * (my_var - my_off)
let my_new_var2=missing(my_new_var, my_new_missing)


Regarding steps (3) and (4) I've got really confused and stuck.

The Ferret 5.80 manual suggests using:

set var/scale=my_scale my_new_var

to set the scale attribute of the scaled variable, and similar for the offset.

However, show commands contains a different syntax:

show commands set/variable
...
SET VARIABLE/TITLE/UNIT/GRID/BAD/DATASET/NAME/SCALEFAC/OFFSET


(Note the "SCALEFAC" spelling.)

Moreover, when I try, no matter which syntax I use, all I get are errors:

yes? set var/scalefac=myscale my_new_var
**ERROR: command syntax: scalefac=myscale
yes? set var/scale=myscale my_new_var
**ERROR: command syntax: scale=myscale
yes? set var/user_scale=myscale my_new_var
**ERROR: unknown command qualifier: user_scale=myscale
yes? set var/nc_scale=myscale my_new_var
**ERROR: unknown command qualifier: nc_scale=myscale


Whereas, the SAY command with RETURN clause seems to work,
with yet different keywords:

yes? say `my_new_var,return=nc_scale`
!-> MESSAGE/CONTINUE 1
1
yes? say `my_new_var,return=user_scale`
!-> MESSAGE/CONTINUE 1
1
yes? say `my_new_var,return=nc_offset`
!-> MESSAGE/CONTINUE 0
0
yes? say `my_new_var,return=user_offset`
!-> MESSAGE/CONTINUE 0
0


Questions:

1. How can I set the scale and offset of the new (scaled) variable?
2. How can I convert it from floating point or double to 2-byte integers?
3. Is this really implemented in Ferret 5.80, i.e. there a way to do what I want using Ferret 5.80?
4. Is there any special function to convert a variable from float to two-byte integer?
5. Is there an example script of how to do all of this?
6. Would it be possible to clarify the meaning, purpose, and differences of all these scale/offset keywords?
(SCALE, SCALEFAC, NC_SCALE, USER_SCALE, and similar for offset)

Thank you.

Gus Correa




[Thread Prev][Thread Next][Index]

Dept of Commerce / NOAA / OAR / PMEL / TMAP

Contact Us | Privacy Policy | Disclaimer | Accessibility Statement