10 - 50 mb Geopotential Heights
                      --   -- -- ------------ -------


Even a casual glance at the 10 mb geopotential height will indicate some 
problems with the upper-level analyses.  In a field that is dominated by wave 
number 1 and 2, the reanalyses shows strong orographic influences.  Attempts 
were made to improve the sigma-to-pressure interpolation with no success.  
The solution may involve either going to a hybrid vertical coordinates, or by 
using a much higher vertical resolution in the stratosphere.  The model uses 
finite differences in the vertical, so a more accurate scheme may reduce the 
"small-scale noise".  Unfortunately tests with a finite-element scheme did not 
show an improvement.



    At 20S, we can draw a line across the Andes:

               points on Gaussian Grid

    Long.      hgt      SPR*   1-PRS  2-PRS  3-PRS

    73.125 W   -539 m   1080    2.92  10.9l  19.76
    71.250 W    -48 m   1019    2.75  10.29  18.65
    69.375 W   2266 m    775    2.09   7.83  14.18
    67.500 W   4183 m    618    1.67   6.24  11.31

    SPR*  surface pressure Jan 1985
    1-PRS pressure of top sigma level
    2-PRS pressure of second from top sigma level
    3-PRS pressure of third from top sigma level



From the above table, we see that (1) the sigma levels have considerable slope 
(in pressure coordinates), (2) the vertical resolution is relatively coarse 
(for the stratosphere), and (3) the Gibbs effect on the surface height. This 
suggests that both the sigma-to-pressure conversion, and dynamics may have 
numerical inaccuracies.