wgrib2: new_grid
Introduction
Beginners are encouraged to read the new_grid introduction.
The new_grid option interpolates the fields to a new
grid.
The default interpolation is bilinear but that can be changed using the
new_grid_interpolation option. This option uses
scalar and vector interpolation as appropriate. In order for the vector
interpolation to work, the vector quantities must be in a (U,V) order.
For example: Z200, U200, V200, Z500, U500, V500 is good. If the data
are not in (U,V) order, the inventory can be sorted into (U,V) order
and the inventory can then be used to control the order of processing.
If the vector quanties are not in (U,V) order, the vector quantites will
not be interpolated. (See example 3 for converting the file into a (U,V) order.)
The option is not part of the default configuration but it included with many distributions.
The interpolation code is written
in fortran and combining fortran and C code can require some work. (gcc/gfortran and
clang/gfortran are already handled by the makefile.) Getting the C and Fortran code to cooperate
requires some systemspecific knowledge and may not be possible in all cases. Consequently
you are on your own in getting the new_grid option installed.
Operations often has to use the new_grid option to produce a large number
of user grids. Fortunately the interpolation can be made embarrassingly parallel. A
portable singlenode solution is described in here. A
multinode solution is possible using MPI and the wgrib2 library.
Caution
The new_grid option works in raw scan mode, so data are not
converted to sn:we order. Conequently options that only work in sn:we order cannot
work at the same time as the new_grid option. Any option
that uses geolocation (ex. rpn, lon)
is incompatible with new_grid.
Winds
Before you do an interpolation, you need to define the wind directions.
Most people want the the V winds to be in the direction of the North Pole.
With a verbose wgrib2 inventory, you will seee winds(N/S). However, some
meteorologists want the V winds to go from grid point (i,j) to (i,j+1).
The corresponding wgrib2 notation is "winds(grid)". See the
new_grid_winds option for more details.
Usage
new_grid_winds W new_grid A B C outfile
W = earth or grid
earth means that the U wind goes eastward, V goes northward
grid means that U wind goes from grid (i,j) to (i+1,j)
which is not eastward in a Lambertconformal or polar stereographic grids
A, B, C are the output grid description
outfile is an output file. The grib2 interpolated records are written in outfile
Grid description format
new_grid ncep grid I outfile I = 2,3,98,126129,170,173,221,230,241,249
(NCEP grid defintions)
T62,T126,T170,T190,T254,T382,T574,T1148,T1534
(NCEP Gaussian grids definitions)
Want more ncep grids? Modify ncep_grids.c
new_grid A B C outfile General format
A grid name with parameters
B usually of the form lon0:nx,dx
C usually of the form lat0:ny:dy
new_grid latlon lon0:nlon:dlon lat0:nlat:dlat outfile latlon grid
lat0, lon0 = degrees of lat/lon for 1st grid point
nlon = number of longitudes
nlat = number of latitudes
dlon = grid cell size in degrees of longitude
dlat = grid cell size in degrees of latitude
new_grid lambert:lov:latin1:latin2:lad lon0:nx:dx lat0:ny:dy outfile Lambert conic conformal
new_grid lambert:lov:latin1:latin2 lon0:nx:dx lat0:ny:dy outfile lad = latin2
new_grid lambert:lov:latin1 lon0:nx:dx lat0:ny:dy outfile latin2 = latini lad = latin1
lov = longitude (degrees) where y axis is parallel to meridian
latin1 = first latitude from pole which cuts the secant cone
latin2 = second latitude from pole which cuts the secant cone
lad = latitude (degrees) where dx and dy are specified
lat0, lon0 = degrees of lat/lon for 1st grid point
nx = number of grid points in X direction
ny = number of grid points in Y direction
dx = grid cell size in meters in x direction
dy = grid cell size in meters in y direction
note: if latin2 >= 0, the north pole is on proj plane
if latin2 < 0, the south pole is on proj plane
new_grid lambertc:lov:latin1:latin2:lad lonc:nx:dx latc:ny:dy outfile Lambert conic conformal with centered position
new_grid lambertc:lov:latin1:latin2 lonc:nx:dx latc:ny:dy outfile like lambert except lonc and latc replace lon0 and lat0
new_grid lambertc:lov:latin1 lonc:nx:dx latc:ny:dy outfile latc, lonc = degrees of lat/lon for center of the grid
new_grid nps:lov:lad lon0:nx:dx lat0:ny:dy outfile north polar stereographic
new_grid sps:lov:lad lon0:nx:dx lat0:ny:dy outfile south polar stereographic
lov = longitude (degrees) where y axis is parallel
to meridian
lad = latitude (degrees) where dx and dy are specified
note: grib1 uses lad = 60N (nps) or 60S (sps)
lad must be 60 (nps) or 60 (sps) (library limitation)
lat0, lon0 = degrees of lat/lon for 1st grid point
nx = number of grid points in X direction
ny = number of grid points in Y direction
dx = grid cell distance meters in x direction at lad
dy = grid cell distance meters in y direction at lad
new_grid gaussian lon0:nx:dlon lat0:ny outfile global Gaussian grid
lat0, lon0 = degrees of lat/lon for 1st grid point
note: lon1 = lon0, lat1 = lat0 + (nx1)*dlon;
nx = number of grid points in X direction
ny = number of grid points in Y direction
ny must be even
dlon = degrees of longitude between adjacent grid points
note: wgrib2 supports regional Gaussian grids, but
the interpolation library doesn't.
new_grid mercator:lad lon0:nx:dx:lonn lat0:ny:dy:latn outfile mercator grid
lad = latitude (degrees) where dx and dy are specified
lat0, lon0 = degrees of lat/lon for 1st grid point
latn, lonn = degrees of lat/lon for last grid point
nx = number of grid points in X direction
ny = number of grid points in Y direction
dx = grid cell distance in meters in x direction at lad
dy = grid cell distance in meters in y direction at lad
note: the mercator grid description is over specified
User must make sure (nx,dy) is consistent with lonn
as well as (ny,dy) is consistent with latn
for v2.0.8 using ip2lib_d (default)
new_grid rotll:sp_lon:sp_lat:sp_rot lon0:nlon:dlon lat0:nlat:dlat outfile
rotated latitudelongitude grid
sp_lon = longitude of the South pole (for rotation)
sp_lat = latitude of the South pole (for rotation)
sp_rot = angle of rotation (degrees)
The grid is defined in the rotated coordinates.
lat0, lon0 = degrees of lat/lon for 1st grid point
nlat = number of longitudes
nlon = number of latitudes
dlon = grid cell size in degrees of longitude
dlat = grid cell size in degrees of latitude
lambert, nps, sps, mercator only support we:sn ordering
latlon, gaussian only support we:sn and we:ns ordering
Finding the Grid Format from Existing File
Sometimes one wants to convert grib files to a common grid. You already
have a sample grib file and you have to figure out the grid format.
For common grids, you can use the perl script
grid_defn.pl
to determine the grid format.
$ grid_defn.pl small.grb2
latlon 0.000000:2:10.000000 20.000000:2:8$
grid_defn.pl does not add a newline at the end of its output.
Not all grids are supported.
Examples
wgrib2 IN.grb set_grib_type same new_grid_winds earth new_grid latlon 100:10:1 30:20:1 OUT.grb
Interpolates from IN.grb to OUT.grb
Uses the same grib packing as in the input file
Makes a 10x20, 1x1 degree latlon grid, lower left corner: 100E 30N
wgrib2 IN.grb set_grib_type same new_grid_winds earth new_grid ncep grid 221 OUT.grb
Interpolates from IN.grb to OUT.grb
Uses the same grib packing as in the input file
Interpolates to NCEP grid 221.
wgrib2 IN.grb set_grib_type same new_grid_winds earth new_grid `grid_defn.pl OLD.grb` OUT.grb
Interpolates from IN.grb to OUT.grb
Uses the same grib packing as in the input file
Interpolates using the grid format of OLD.grb (1st record)
In this example, U and V are not in the required order. This
shows a sorting to the required order for new_grid to work.
wgrib2 201201.A  sed e 's/:UGRD:/:UGRDa:/' e 's/:VGRD:/:UGRDb:/'  \
sort t: k3,3 k5,8 k4,4  \
wgrib2 201201.A i new_grid_winds earth new_grid ncep grid 2 201201.A.grd2
The first line creates an inventory with new variable names: UGRD > UGRDa and VGRD > UGRDb
The second line sorts the inventory so that UGRDb follows UGRDB.
The third line regrids the file, with the order of processing controlled by the inventory.
Type of Interpolation
The IPOLATES (either iplib or ip2lib) library supports a number of interpolation schemes including bilinear (default),
bicubic, neighbor and budget. The wgrib2 was tested with the spectral interpolation but
that option was removed because it greatly increased the code size for a minor option.
The interpolation method can be selected by using the new_grid_interpolation
option before the new_grid option. Some of the interpolation
options need numeric parameters which are set by the
new_grid_ipopt option. IPOPT is defined in the iplib library documentation.
You can use different interpolations for different variables. For example, a
bilinear interpolation of soil or vegetation type is meaningless. So
nearest neighbor interpolation is used instead.
wgrib2 IN.grb new_grid_winds earth \
new_grid_interpolation bilinear \
if ":(VGTYPSOTYP):" new_grid_interpolation neighbor fi \
new_grid latlon 0:360:1 90:181:1 OUT.grb
line 2: set default interpolation to bilinear
line 3: if VGTYP or SOTYP then set the interpolation to nearest neighbor
line 4: do the interpolation
Comments (4/2018): When you convert from a high resolution grid
to a lower resolution grid, you have to be consider changing from
the default interpolation (bilinear) to a budget interpolation.
The budget gives a better estimate of the cell average (the default
is 25 bilinear interpolations).
Changing from gridrelative to Earthrelative winds and vice versa
Most NCEP grib files use gridrelative winds. If you want to convert to Earthrelative winds
or gridrelative winds, you can use the new_grid option.
To Earth relative:
wgrib2 IN.grb set_grib_type same new_grid_winds earth new_grid_interpolation neighbor \
new_grid `grid_defn.pl IN.grb` OUT.grb
To Grid relative:
wgrib2 IN.grb set_grib_type same new_grid_winds grid new_grid_interpolation neighbor \
new_grid `grid_defn.pl IN.grb` OUT.grb
"set_grib_type same" preserves the grib packing or compression
"new_grid_interpolation neighbor" should be faster than the default bilinear
`grid_defn.pl IN.grb` returns the grid definition of the first grib message in IN.grb
The limitations of the above command are
 IN.grb can only have one grid type
 OUT.grb will have any submessages converted into messages
Changes from copygb
 copygb default vectors: UGRD/VGRD
 wgrib2 default vectors: depends on version of wgrib2. See new_grid_vectors.
 copygb can have vectors in any order
 wgrib2 must have V follow U for vectors pairs
 copygb has bilinear, bicubic, nearest neighbor, budget, neighbor budget, and spectral interpolations.
 wgrib2 has bilinear, bicubic, nearest neighbor, and budget interpolations.
 copygb uses fixed Earth's radius
 wgrib2 uses Earth's radius based on grib message
 wgrib2 doesn't have merging, mapthreshold or map files
 copygb by default, ignores the binary scaling and preserves decimal scaling
 wgrib2 by default, preserves binary and decimal scaling
 copygb does grib1.
 wgrib2 does grib2.
Speed: Interpolation Weights
The first step of the new_grid interpolation is
to calculate the interpolation weights. (Each grid point on the new grid
is a weighted average of a small set of the old grid points.)
To save time for future interpolations, the last set of weights is saved.
Consequently interpolation is fastest when the input and output grids remain
don't change. While one can have multiple new_grid options on
the command line, it is not recommended because the caching of the weights
wouldn't work and weights would have to be recalculated every time.
Installation
1) set USE_IPOLATES to 1 in the makefile This is the default (5/2018).
2) environment variable $FC must be set to the fortran90 compiler, ex. export FC=gfortran
3) If $FC is not gfortran, openf95, or xlf_r, then
modify the makefile for the new fortran compiler
modify New_grid.c for the new fortran compiler
Note: icc and pgc are incompatible with the jasper library and are not supported
4) now you are ready to make
Converting from WE:SN to WE:NS Grids
Many of wgrib2 grib2 writing options will write the grid
in WE:SN order. This natural because geolocation is
only enabled when the internal grids are in WE:SN order.
However, some codes need the grid in WE:NS order. To
convert a grib file from WE:SN order to WE:NS order, the
simplest way is to use new_grid. Lat0 and lon0 need
to be lat/lon of the top left corner of the grid. Dlon
will a positive number and dlat will be negative.
If you want to be tricky, you can do a variation of the
"NDFD work arounds" technique. It will be faster and
more generic.
Quilting tiles  Merging files
Yes, it can be done using new_grid, bin, import_bin rpn, and
the grib_out options. Yes, I have done it.
Limitations of the iplib library
The iplib library has its limitations.
Grids types used by NCEP get supported and others don't. The library has a grib1 interface, so
new features of grib2 are not supported and the precision of the grid parameters
are limited to the grib1 values. For example, latitude and longitude
values are limited to millidegrees instead of microdegrees. Note: during
the installation of ip2, I noticed that the longitude was reduced when
reducing the range to [0,360). (longitude = amod(longitude + 3600.0, 360.0))
For single precision numbers, this reduces the precision to about two digits after
the decimal place.
 not all grid types are supported by iplib
 only common grids are supported by the wgrib2 "wrapper" for iplib
 latitude, longitude values are nomially in millidegrees (affects interpolation)
 Single precision reals are used. The effective precision of the
longitudes is to hunderdth of a degree.
 only grib1 scan order are supported (i.e., WE:SN and WE:NS)
 NDFD/Glahn scan order is not supported (i.e., WEEW:SN)
 Earth is assumed to be spherical
 Lambert conformal: LatD must follow grib1 conventions
 Polar Stereographic: LatD must be 60 latitude (grib1 convention)
 nx, ny, npts must be ≤ 2147483647. (Grib2 standard is 4294967295.)
Limitations of the ip2lib_d library (wgrib2 v2.0.8+)
As of wgrib2 v2.0.8, ip2lib_d will be the default interpolation library.
Ip2lib_d is the doubleprecision grib2interface IPOLATES library. The
old ip2lib will remain a compile time option. The ip2lib_d is the new
default IPOLATES library because is supports rotated latlon grids (Grid
Definition Template 3.1) which will used in the regional FV3 model at
NCEP. Canada is planning to release model forecasts using this grid,
and the Europeans are already using this grid for their regional models.
The speed of the single precision iplib was originally faster than
the double precision ip2lib_d. To remove this objection from converting
to the new library, a effort was made to increase the OpenMP threading.
Consequently the run time of ip2lib_d will be similar (bilinear) to much
faster (budget) than iplib in a multiprocessor environment.
 supports more grids than iplib including WMO defined rotated latlon.
 latitude, longitude values are in microdegrees (affects interpolation)
 Double precision reals are used. (Wgrib2 stores is grid point values
as single precision.)
 grib1 polar stereographic projection, LaD = 60N or 60S has been removed
 NDFD/Glahn scan order (i.e., WEEW:SN) will be supported unlike iplib
 staggering will be supported unlike iplib
 Some interpolations support an ellipical earth.
 Lambert conformal: LatD must follow grib1 conventions
 nx, ny, npts must be ≤ 2147483647. (Grib2 standard is 4294967295.)
This is a limitation of compiling using 4 byte integers.
Future changes
The ip2lib_d library will replace iplib as the default (v2.0.8).
NDFD work arounds
NDFD files are often written (WEEW):SN order. This means that the odd rows
are in WE order and the even rows are EW order. The rows go from south to north.
If you are using wgrib2 that is using iplib, the simplest solution
to convert the grid to WE:SN order is
Version 0
 for data in input=WEEW:SN scan order (wgrib2 IN scan)
 read data, change order scan order of data, change flag table 3.4, save data
 wgrib2 IN.grb rpn alt_x_scan set table_3.4 64 grib_out OUT.grb
Version 1
 Read trick 55 in wgrib2 tricks
Version 1
A variation of the previous trick can be used to put data in to (WEEW):SN packing.
See also:
new_grid introduction,
new_grid_interpolation,
new_grid_winds,
new_grid_vectors,
lola,
bin,
import_bin,
rpn,
grib_out,
new_grid multicore usage,
