Re: ECMWF operational analysis

From: Dennis Shea <shea_at_nyahnyahspammersnyahnyah>
Date: Wed, 10 May 2006 11:38:12 -0600 (MDT)


The following is a response to a question posed
>Date: Thu, 23 Mar 2006 13:15:26 +0100
>Hi all,
>[snip] but I couldn't find
>anything to solve my problems, anyway,
>I got an ECMWF operational analysis via the MARS
>archive, (only Z,U,V,Q-rel.hum.-,T for 4 press levels),
>In my ncl script I print out the variables and
>the isobaric levels:

The ECMWF operational analysis referred to is on a *very*
high resolution grid: NLAT=800, MLON=1600.

The file name is "ana_yest.grb". The contents of the
file may be viewed at:

One variable, Q_GDS4_ISBL (specific humidity) is on
the 800x1600 grid.

   float Q_GDS4_ISBL ( lv_ISBL0, g4_lat_4, g4_lon_5 )
                          4 800 1600
      long_name : Specific humidity
      units : kg kg**-1
Thus, Q_GDS4_ISBL can readily be read and plotted via NCL. EG: See:

All other variables are in "complex coefficient" form. 
A sample complex coefficient variable [temperature] looks like
   float T_GDS50_ISBL( lv_ISBL0, real_imaginary, g50_lat_2, g50_lon_3 )
                          4          2              800        800
      long_name :	Temperature
      units :	K
The data fields must be reconstructed [synthesized] 
prior to plotting.
In all previous occurrences,
NCL's spherical harmonic synthesis functions and procedues
have always been able to reconstruct the variables. In fact,
they exactly agree with ECMWF's AMOS software. However,
when the standard approach was used on these 800x800 coefficients,
the generated fields contained numerical exceptions [NaNs].
The 'problem' is that the standard recursive methods 
used for the Legendre polynomials are not converging.
Currently, it looks like nlat up to 646 will converge,
        nlat = 646
        Dimensions and sizes:   [lat | 646] x [lon | 1292]
        Temperature: min=232.126   max=303.634
        nlat= 648           
        Dimensions and sizes:   [lat | 648] x [lon | 1296]
        Temperature: min=nan0x7fffffff   max=nan0x7fffffff
Thus, currently the full 800x1600 grids can not be 
generated from the full set of complex coefficients. 
In practice, the very high numbered coeffients are near zero 
so the additional information is not significant.
A script that reads/synthesizes/plots all variables 
on the file is at
Q is plotted at 800x1600. The variables generated
from spherical harmonics are synthesized to a 640x1280 grid.
An example for temperature is at:
If anyone has questions regarding the script, please
feel free to contact me directly.
Dennis Shea
ncl-talk mailing list
Received on Wed May 10 2006 - 11:38:12 MDT

This archive was generated by hypermail 2.2.0 : Wed May 17 2006 - 09:57:59 MDT