wgrib2: rpn
Introduction
The rpn option runs a reverse polish notation
(RPN)
calculator. Having a builtin calculator is quite handy. We use it to convert
units (ex. geopotential to geopotential meters, accumulations to rates),
compute simple quantities (net flux from downward and upward fluxes),
and even compute the plant hardiness index from the 2 m temperatures. The goal
of the calculator is to reduce the need to write simple grib programs that
do simple calculations.
The "hardware" of the rpn calculator consists of 20 registers
and a stack (10 entries deep). (Wgrib2 prior to 2.0.6 has 10 registers.)
Stack entries and registers are arrays rather floating point numbers on
your storebought calculator.
The conceptual model of the >rpn calculator is the
grid values array is the top of the stack. The calculator has
a statck that is 10 entries deep and 20 registers. When
ever you enter rpn mode, the stack is cleared except for the
top of the stack. The registers are only clears when you
"turn on" the calculator; that is start the wgrib2 utility
or the first call to wgrib2 if you are using calling wgrib2
through wgrib2api or pywgrib2.
To save the calculations, you can save them in a register
or write them out by grib_out, bin, ieee, text, etc.
Callable wgrib2, wgrib2api
Callable Wgrib2 and wgrib2api use the rpn registers to transfer
gridded data between the calling program and the wgrib2 subroutine. A Fortran or C
program can read and write the wgrib2's RPN registers. For example, if a program
wants to write a grib2 file, it would first place the grid values into a register.
The calling program would then call the wgrib2 subroutine with instructions to
read a template, replace the grid values with the register values, and then
write a grib message.
Implementation Details
The size of a registers may differ as the size of grids can also vary in a grib file.
However the size of the register has to match the size of the stack entry in order
"recall" the register.
Wgrib2 always reads a grib message before processing it using commmands like rpn.
This sets the size of the data array. Thus the size of the stack entries is always
the size of the grib message that is being processed.
The conceptual moddel is the data array (grid values) is the
top of the stack. The implmentation is that the data array is copied
to the top of the stack when rpn is run, and the top of the stack is
copied to the data array when rpn is finished. An error message
will occur if the top of the stack is empty when rpn finishes.
Limitations of rpn
The rpn option was designed for
simple calculations. For more complicated calculations, you
should use a real programming language. You can do the
calculations in another step and then import the results
using one of the various import_* options. You can
use wgrib2api (Fortran and C) to read the data into a Fortran
or C program, do the calculation and then write the data out
using wgrib2. Finally you can use python and one of the pywgrib2
packages to do the calculation using Python, numpy and
pywgrib2.
Uses
 change of units when importing data (gribifying data)
 computations: ex, U,V > wind speed, wind direction, potential temperature
 merging data
 complex masking of data
 changing units before writing text/ieee files
 removing extreme data values
 finding min and max values
 finding the globally averaged precipitation
 comparing fields
Usage
rpn "A:B:C:..."
A,B,C,.. = number, rpn function, or rpn operator
Operators and Functions:
Pop X, Push Fn(X)
 abs: absolute value
 acos: arc cos, [0, pi] radians
 alt_x_scan: changes alternate x scanning to regular x scanning and vice versa
 asin: arc sin, result is [pi/2, pi/2] radians
 atan: arc tan, result is [pi/2, pi/2] radians, see atan2
 abs: absolute value
 ceil: smallest integer >= X
 cos: cosine, argument in radians
 exp: e^X
 floor: largest integer <= X
 ln: natural logorithm
 raw2: convert from input scan order to output scan order
 sin: sine, argument in radians
 smth9g: smth9 for global fields
 smth9r: snth9 for regional fields
 sq: X*X
 sqrt: square root
 tan: tangent, argument in radians
 xave: for nxny grids, averages in the x direction (normally zonal mean)
 xdev: for nxny grids, remove x average (normally deviation from zonal mean)
 yrev: for nxny grids, changes we:sn to we:ns and vice versa
 1/x: 1/X
 2raw: convert from output scan order to input scan order
Pop Y, Pop X, push Fn(X,Y)
 +: push X+Y
 : push XY
 *: push X*Y
 /: push X/Y
 <: push X < Y (1/0 if true/false)
 <=: push X <= Y (1/0 if true/false)
 ==: push X == Y (1/0 if true/false)
 !=: push X != Y (1/0 if true/false)
 >=: push X >= Y (1/0 if true/false)
 >: push X > Y (1/0 if true/false)
 atan2: push arctan(X/Y), result is [pi, pi] radians, see atan
 pow: push X**Y (X^Y)
 mask: if (Y != 0) push(X) else push(UNDEFINED)
 max: push max(X,Y), for logical values, max is the same as OR
 merge: if (Y != UNDEFINED) push(Y) else push(X)
 min: push min(X,Y), for logical values, min is the same as AND
Note: an operation involving an UNDEFINED is UNDEFINED
Stack Operators:
 clr, clear the stack (stack is emptied)
 dup, duplicate the top of the stack
 pop, remove the top of the stack
 exc/swap, exchange the top 2 stack entries
Register Operators: (note: CW2 v2.0.6+ uses registers 7,8,9 prior versions 0,1,2)
 clr_I, clear register I, I=0,1..,9 (19 for v2.0.6+)
 rcl_I, push register I on top of stack, I=0,1..,9 (19 for v2.0.6+)
 sto_I, save top of stack in register I, I=0,1..,9 (19 for v2.0.6+)
 rcl_lat, push latitudes onto the top of the stack (degrees)
 rcl_lon, push longitudes onto the top of the stack (degrees)
 sto_lon, save top of stack as longitudes (degrees) (wgrib2 v3.0.0+)
 sto_lat, save top of stack as latitudes (degrees) (wgrib2 v3.0.0+)
 note: latitudes and longitudes are double precision values, the stack is single precision
Variables and Constants: put on the top of the stack
 number number = floating point or integer number like 0, 10.1, 1.23e4
 days_in_ref_month number of days in the month for the reference date (conversion between monthly acc. and rates)
 days_in_verf_month number of days in the month for the verification time (conversion betwee monthly acc. and rates)
 pi 3.1415....
 rand random number uniformly distributed between 0 and 1, each grid point has a different random number
Printing Operators:
 print_corr, write cosine weighted spatial correlation, data[TOP] data[TOP1]
 print_max, print_min, data[TOP]
 print_rms, write cosine weighted RMS, data[TOP]data[TOP1]
 print_diff, write cosine weighted difference, data[TOP1]data[TOP]
 print_ave, write cosine weighted average, grid_ave(data[TOP]*cos(grid))/grid_ave(cos(grid))
 print_wt_ave, write weighted average grid_ave, data[TOP] is the weighting
print grid_ave(data[TOP]*data[TOP1])/grid_ave(data[TOP])
Example 1
The standard units of grib temperature is K but you want the text output in Celcius.
$ wgrib2 a.grb match ":TMP:850 mb:" rpn "273.15:" text C.dat
Fahrenheit is easy too (F = (K273.15)*9/5+32).
$ wgrib2 a.grb match ":TMP:850 mb:" rpn "273.15::9:*:5:/:32:+" text F.dat
Example 2
Suppose you want to limit the relative humidity values to 100. This example only
affect the RH fields. All submessages will be converted into messages.
$ wgrib2 a.grb if ":RH:" rpn "100:min" fi grib_out out.grb not_if ":RH:" grib out.grb
Example 3
Suppose that you wanted the 500 to 1000 mb thickness, and the file only contained
one field of Z1000 and one field of Z500.
$ wgrib2 IN.grb match ":HGT:" match ":(5001000) mb:" \
if ":500 mb:" rpn sto_1 fi \
if ":1000 mb:" rpn sto_2 fi \
if_reg "1:2" \
rpn "rcl_1:rcl_2::clr_1:clr_2" \
set_var THICK set_lev "5001000 mb" \
set_grib_type c3 grib_out OUT.grb
line 1: only process the HGT at 500 and 1000 mb which save processing time
line 2: store HGT at 500mb in register 1
line 3: store HGT at 1000mb in register 2
line 4: if (register 1 and register 2 have values then
line 5: calculate the thickness: reg_1  reg_2
line 6: set variable type to THICK, and level to "5001000 mb"
line 7: write out the WIND data to a grib file with complex compression
Note: this is a very simple script and that doesn't check the matching
date code, grid type, etc.
Example 4
Write out the 500 mb wind speed.
$ wgrib2 IN.grb match ":[UV]grd:500 mb:" \
if ":UGRD:" rpn "sto_1" fi \
if ":VGRD:" rpn "sto_2" fi \
if_reg 1:2 \
rpn "rcl_1:sq:rcl_2:sq:+:sqrt:clr_1:clr_2" \
set_var WIND \
grib_out out.grb
line 1: only process the U and V at 500 mb
line 2: store U 500mb in register 1
line 3: store V 500mb in register 2
line 4: if (register 1 and register 2 have values then
line 5: calculate the wind speed: sqrt(reg_1**2 + reg_2**2)
line 6: set variable type to WIND (wind speed)
line 7: write out the WIND data to a grib file
Note: this is a very simple script and that doesn't check the matching
date code, grid type, etc.
Note: there are options to calculate wind speed and wind direction
Example 5
Suppose someone made a mistake and the latent heat flux (LHTFL) had the wrong sign. RPN to the rescue.
$ wgrib2 IN.grb match ":LHTFL:" rpn "1:*" grib_out new_lhtfl.grb
The file, new_lhtfl, only contained the LHTFL fields. You duplicate the file with the
fixed LHTFL fields by
$ wgrib2 IN.grb if ":LHTFL:" rpn "1:*" fi grib_out new.grb
It would be faster if you only compressed the LHTFL fields. (grib uses the
original compressed data and grib_out uses the "data" register.)
$ wgrib2 IN.grb set_grib_type jpeg \
not_if ":LHTFL:" grib new.grb if ":LHTFL:" grib_out new.grb
If both the latent and sensible heat fluxes needed a sign reversal, you could do,
$ wgrib2 IN.grb if ":(LHTFLSHTFL):" rpn "1:*" fi grib_out new.grb
Example 6
If you want to set certain values to undefined, you define a mask and then
apply the mask. In this example, values below 20 are set to undefined.
$ wgrib2 a.grb rpn "dup:20:>=:mask" grib_out set_grib_type c3 new.grb
The RPN calculator is used:
dup the data is duplicated
20 20 is pushed on the stack
>= test data >= 20, top of stack is 1/0 depending on test >= 20
mask apply mask to the data
set_grib_type c3 sets the grib compression to complex3
grib_out new.grb writes a grib message using the decoded data
Don't forget to enclose the argument to rpn in quotes because the shell can do unexpect things.
Printing operators
print_corr write cosine weighted spatial correlation R(TOP1), R(TOP)
print_max write max(R(TOP)) to stdout
print_min write min(R(TOP)) to stdout
print_rms write cosine weighted RMS(R(TOP1)R(TOP))
Example 7: Merging
Suppose that we have a nested model, we have a low resolution TMP2m from the
the outer model and a highresolution TMP2m from the nested model. Now we
want a field that uses the TMP2m in the nestedmodel domain and the TMP2m from
the outer model elsewhere. To do this, you need to convert both fields to a
common grid. Then you use "rpn merge". Make sure that both domains are
contained in the common grid as this is a requirement of the interpolation library.
wgrib2 OUTER.T2m new_grid_winds earth new_grid A B C A1.grb
wgrib2 NESTED.T2m new_grid_winds earth new_grid A B C A2.grb
wgrib2 A2.grb rpn sto_1 import_grib A1.GRB rpn "rcl_1:merge" \
grib_out MERGED.T2m
Example 8: Land Mask
The file mask.grb contains the values 0 for water, 1 for land and
2 for sea ice. I wanted a small file with 0 for water+seaice and 1 for land.
wgrib2 mask.grb rpn "1:==" set_scaling 0 1 set_grib_type c1 grib_out land.grb
rpn "1:==" if grid value is 1, the new value is 1 else 0
set_scaling 0 1 1 bit for storing the grib values
set_grib_type c1 complex packing 1 is good for long runs of the same value.
The file sizes 16 bits/point 16 bits precision simple packing (mask.grb)
0.7 bits per point 16 bits precision complex packing 1 (mask.grb)
0.2 bits per point complex packing 1 (land.grb)
grid size: 131K points, land.grb is 3331 bytes
Example 9: Totaltotal index
An example of calculating the dewpoint and
totaltotal index is more involved. Using an online infix to postfix (reverse polish)
calculator is helpful.
Example 10: global precipitation
The model has the precipitation in the variable PRATE
which has units of mm/sec (assuming 1 gm H2O = 1cc). Suppose
I wanted the globally averaged precip in the nonmetric unit
of mm/day. It's one command away
wgrib2 gfsfile match PRATE s rpn "86400:*" stats
Changes
Wgrib2 v3.0.0+ adds error chechking to floating point values,
wgrib2 prior to 3.0.0
wgrib2 IN.grb rpn "2cars:+" grib_out OUT.grb
adds 2 to the grid point values as atof("2cars") returns 2.0
wgrib2 v3.0.0+
wgrib2 IN.grb rpn "2cars:+" grib_out OUT.grb
will result in an error message as "2cars" is not a legal floating point value
Comments
Warning: Reverse Polish notation can cause headaches if you try something
too complicated. An infix > postfix calculator is the suggested
remedy. Another approach is to use pywgrib2_s, pywgrib2_xy, or pywgrib2_lite.
The rpn option is a piece of easy to
understand and modify code (RPN.c). If you want to add
a specialized function (ex. wind chill calculation), you many
consider adding it to the RPN calculator. The another method
is to code your calculation in python and use pywgrib2_s,
pywgrib2_xy or pywgrib2_lite.
Why an RPN calculator? Well, wgrib2 is heavily influenced by the stack language Forth.
It's only natural that the calculator would be based on reverse
Polish notation.
See also:
if,
if_reg,
fi,
grib_out,
rpn_rcl,
rpn_sto,
