Limited precision on x/y increments for interpolation
Added by Jordi Mercader about 5 years ago
Dear all,
So far, I've been using an old version of CDO (1.5.4) to perform a bilinear interpolation between lat-lon grids. It allowed me to set an X,Y increment on the grid description file with up to 8 decimals places that were taken into account:
[meteo@recercam1 opt]$ cat pdc.09.grid
[..]
gridtype = lonlat
gridsize = 68120
xname = lon
xlongname = longitude
xunits = degrees_east
yname = lat
ylongname = latitude
yunits = degrees_north
xsize = 260
ysize = 262
xfirst = 0.147870
xinc = 0.012324000
yfirst = 40.512275
yinc = 0.009114000
So, when we ran
cdo remapbil,pdc.09.grid out.grib pdc.grib
the output grib had effectively this X/Y increment. We checked it with grib_dump:
/opt/eccodes-2.12.0/bin/grib_dump pdc.grib
[...]
jDirectionIncrementInDegrees = 0.00911494;
iDirectionIncrementInDegrees = 0.0123243;
[...]
However, when using CDO v1.9.5, we can still use the same descriptor file but only 3 decimal places are considered:
/opt/eccodes-2.12.0/bin/grib_dump pdc.grib
[...]
jDirectionIncrementInDegrees = 0.009;
iDirectionIncrementInDegrees = 0.012;
[...]
and, consequently, the new grid has different interpolated values when compared to the old version.
Do you know if there is a way to take into account more than 3 decimal places for X/Y increments when performing an interpolation? Thank you very much!
Replies (3)
RE: Limited precision on x/y increments for interpolation - Added by Ralf Mueller about 5 years ago
grib encodes double values with lossy compression. Avoid lengthy floating point numbers when working with grib1. You could use grib2, this works better:
<ram@melian:~/Downloads> % cdo -f grb2 -setname,orog -topo,target.grid t.grib2 cdo(2) topo: Process started cdo(2) topo: cdo setname: Processed 68120 values from 1 variable [0.06s 50MB] <ram@melian:~/Downloads> % cdo griddes t.grib2 # # gridID 1 # gridtype = lonlat gridsize = 68120 xsize = 260 ysize = 262 xname = lon xlongname = "longitude" xunits = "degrees_east" yname = lat ylongname = "latitude" yunits = "degrees_north" xfirst = 0.14787 xinc = 0.012324 yfirst = 40.512275 yinc = 0.009114 cdo griddes: Processed 1 variable [0.12s 66MB]For the target grid I used your input
% cat target.grid gridtype = lonlat gridsize = 68120 xname = lon xlongname = longitude xunits = degrees_east yname = lat ylongname = latitude yunits = degrees_north xsize = 260 ysize = 262 xfirst = 0.147870 xinc = 0.012324000 yfirst = 40.512275 yinc = 0.009114000
Whereas when I use grib1:
% cdo griddes t.grib # # gridID 1 # gridtype = lonlat gridsize = 68120 xsize = 260 ysize = 262 xname = lon xlongname = "longitude" xunits = "degrees_east" yname = lat ylongname = "latitude" yunits = "degrees_north" xfirst = 0.148 xinc = 0.0123243243243243 yfirst = 40.512 yinc = 0.00911494252873563
hth
ralf
RE: Limited precision on x/y increments for interpolation - Added by Ralf Mueller about 5 years ago
I use cdo-1.9.6 compiled with eccodes for grib2 support
RE: Limited precision on x/y increments for interpolation - Added by Jordi Mercader about 5 years ago
Thank you Ralf for your advice. I've tried it with grib2 and, as you commented, it keeps all the decimal places.