BRAMS
Minimal Instalation
1. Introduction
The BRAMS minimal instalation don't use some libraries like:
a) GRIB2 (WGRIB2): used to read GFS/NCEP Model's analisys directly;
b) NETCDF4 with CDF5: used to read GEOS/NASA Model's analisys directly;
c) HDF5 : used to read some fixed data like NDVI
The minimal instalation may be used if you're reading GRADS data analisys
generated using data from global model in grads, data fixed using bin or
grads data or the native format from BRAMS called vfm format generated by
dprep software.
2. Prerequisites:
The BRAMS must be compiled using fortran compiler. The Model was tested
using GNU (Gfortran&GCC), Intel compiler(Ifort,icc), Portland (pgf90, pgcc),
IBM, NEC and Cray compilers.
BRAMS works only in parallel mode. One can run the model using a single
processor/core, but must to do it using MPI with the MPIRUN command.
If you don't have a MPI library installed we recommend download and
install the last version of MPICH stable release.
Also take care to choose the correct version to your OS.
example:
> cd ~/install
> wget http://www.mpich.org/static/downloads/3.3.2/mpich-3.3.2.tar.gz
> tar -zxvf mpich-3.3.2.tar.gz
> cd mpich-3.3.2
> ./configure -disable-fast CFLAGS=-O2 FFLAGS=-O2 CXXFLAGS=-O2 FCFLAGS=-O2 \
--prefix=/opt/mpich3
> make
> sudo make install
NOTE: If Your system uses another compile than GNU use it in configure of
MPICH.
3. Getting the Model:
The BRAMS ins under version control using subversion. The stable and beta test
releases are in INPE/CPTEC ftp server. To get the last test realease please
> wget http://ftp.cptec.inpe.br/pesquisa/bramsrd/BRAMS/releases/beta/BRAMS_GEOS.tgz
4. Unpacking & Compiling the Model:
Unpack the model:
> tar -xzvf BRAMS_GEOS.tgz
go to build folder:
> cd BRAMS_GEOS/build
Run a Makefile configure:
> ./configure --program-prefix=BRAMS_5.4 --prefix=/home/<user> --enable-jules \
--with-chem=RELACS_TUV --with-aer=SIMPLE --with-fpcomp=/opt/mpich3/bin/mpif90 \
--with-cpcomp=/opt/mpich3/bin/mpicc --with-fcomp=gfortran --with-ccomp=gcc
Notes:
a) --prefix=/home/<user> - You must choose your <user> instalation area;
b) In example above we use mpich3 and gfortran (GNU) instalation;
c) In example we use the mpi installed in /opt/mpich area;
Make and Make install the model:
> make
> make install
Note: The binary executable will be installed in bin folder created under
--prefix you specified. This will be your "run" area.
5. Getting fixed data and tables
The model uses some fixed files like topography, vegetation, NDVI, etc.
You must get the data and put it in a folder to be readed by model.
5.1 Getting table data:
Get the table data and put it in bin folder created in install:
> cd /home/<user>/bin
> wget http://ftp.cptec.inpe.br/pesquisa/bramsrd/BRAMS/files/bramsTablesRev2000.tgz
> tar -xzvf bramsTablesRev2000.tgz
Note: the table data size is 405MB
5.2 Getting the fix data:
> wget http://ftp.cptec.inpe.br/pesquisa/bramsrd/BRAMS/files/datafix.tar
> tar -xzvf datafix.tar
Note: the datafix data size is 3.5GB
All the data inside datafix is zipped using bzip2. Please, bunziped all the
data in all sub-folders.
6. Getting the Initial and boundary conditions:
The model needs initial and boundary conditions to run. The data must be
obtained in INPE/CPTECs ftp site:
Created a datain subfolder inside Your bin folder:
> mkdir datain
> cd datain
Get all data in http://ftp.cptec.inpe.br/pesquisa/bramsrd/BRAMS/data/GRADS/20200515/
for test (gra & ctl data) for day 15May2020.
You can get data for just one day of forecast:
from IC2020051500.* to IC2020051600.* and SM.GEOS.202005150000.*
or until 10 days of forecast.
7. The tmp folder.
Brams uses Jules Models for surface. Jules needs a temporary directories for
run. Please, make it inside your bin folder and export the tmp folder before
run the model:
> cd /home/<user>/bin
> mkdir tmp
> export TMPDIR=./tmp
8. The NAMELIST RAMSIN
The BRAMS uses by default a namelist called RAMSIN. We disposed a minimal
RAMSIN to be modified by user, get it in ftp area:
> wget http://ftp.cptec.inpe.br/pesquisa/bramsrd/BRAMS/files/RAMSIN_MINIMAL
Please, lets talk about some RAMSIN important variables:
* RUNTYPE = 'MAKESFC': The BRAMS is executed in 3 phases: MAKESFC, MAKEVFILE and
INITIAL. The first to phases prepares the model for run. They Creates fix data
and data in the area of interest. You must run the MAKESFC and MAKEVFILE using just
1 processor (or core). The INITIAL is the integration in time phase. Read the
previous data from 2 previous phases and go forward. The INITIAL phase may be
executed with the maximum of processors possible (it depends os area configuration).
* TIMMAX = 24., : define how long the forecast will be done;
* IMONTH1 =05, ! Month
* IDATE1 =15, ! Day
* IYEAR1 =2020, ! Year
* ITIME1 =0000, ! GMT of model TIME = 0.
The variables above are used to define the initial integration time.
* NGRIDS = 1, : Number of grids to run version 5.x of BRAMS only
admits 1 grid**
* NNXP = 80, 318, 154, ! Number of x gridpoints
* NNYP = 80, 346, 102, ! Number of y gridpoints
* NNZP = 35, 38, 38, ! Number of z gridpoints
* NZG = 7, ! Number of soil layers
* NZS = 1, ! Maximum number of snow layers
* DELTAX = 10000., ! X grid spacing
* DELTAY = 10000., ! Y grid spacing
* DELTAZ = 100., ! Z grid spacing (set to 0. to use ZZ)
* DZRAT = 1.1, ! Vertical grid stretch ratio
* POLELAT = -21.16, ! Latitude of pole point
* POLELON = -44.93, ! Longitude of pole point
* CENTLAT = -21.16, ! Latitude of grid center
* CENTLON = -44.93, ! Longitude of grid center
The variables above define the "size" of area (80 x 80 points), the vertical
number of levels (35) and other levels, The grid spacing in meters
(in this case 10 by 10 km), the initial z level (100m) and a ratio from
levels above the first level. CENTLAT and CENTLON are the center of domain
area en usually are the same of POLELAT and POLELON. Changes in POLES cause
geographical distortion of area.
NOTES:
i. IF YOU CHANGE ONE OF THE VARIABLES ABOVE YOU MUST RUN THE 3 PHASES OF
MODEL AGAIN!
ii. The size defined by default is small enough to run in a ordinary laptop
iii. In case the test with same size of the area/time/date you can only run
the INITIAL phase.
* DTLONG = 60., : Coarse grid long timestep. This is one important value!!!
This variable inform to model integration the delta T used to advance in time.
Small numbers is best but the model will takes a lot of time to end.
Large number may cause a model instability and crash. If You are using
DYNCORE_FLAG = 0 (leapfrog method) pay attention in SSCOURN value printed in
screem when models run, the value must be close to 1. If You are using
DYNCORE_FLAG = 2 (Runge-Kupta) the value may be close to 4. When You run
the model and it crash with blow up of courant number may be necessary
decrease the DTLONG.
NOTE: IF You change the DELTAX or DELTAY the value of DTLONG must be change
accordingly!
FRQANL = 3600., : Time between outputs for forecast. In this case 1 hour.
In some computer systems this time is relevant because are I/O operations
involved. You can test a model without I/O by setting IPOS = 0 and
IOUTPUT = 0. By default we use only pos-processed grads file and turn off
IOUTPUT. For IPOS = 2 the output will be the variables listed in $POST section
(close to end of RAMSIN) and the files output are defined by GPREFIX. In case
GPREFIX = './dataout/forecast'
9. Running the Model:
You must run the 3 phases of Model.
Phase #1) Edit RAMSIN and change RUNTYPE to 'MAKESFC'.
Then submmit the run with only one processor (or core):
mpirun -np 1 brams-5.4 -f RAMSIN_MINIMAL
Pay attention to the output screen. Some erros may occur for a lot of reasons.
Check it carefully in case of problems.
Phase #2) Edit RAMSIN and change RUNTYPE to 'MAKEVFILE'.
Then submmit the run with only one processor (or core):
mpirun -np 1 brams-5.4 -f RAMSIN_MINIMAL
Phase #3) Edit RAMSIN and change RUNTYPE to 'INITIAL'.
Then submmit the run with a lot of processors (or cores):
mpirun -np 40 brams-5.4 -f RAMSIN_MINIMAL
NOTE: The number of cores You run depends of the machine architecture and the
model's grid defined. If You used more cores than the model can support for one
grid You must tune the grid. In general the problem accour when the number of
grids columns by core is less then halo of MPI communication. Small area with
little colunms do not be used with many cores. The model will notice You when
the problem is present.
To see the models forecast just look inside the output GPREFIX folder to see
the GRADS files generated.
10. The dump in screen
The BRAMS model print some important information during the run of each phase.
Check it for found errors or mistakes in configuration. One important information
are the lines:
=== Initial timestep info: ngrid, nndtrat, nnacoust, dtlongn, sscourn, sspct ===
1 1 4 60.000 4.204 1.000
The sscourn number (4.204) indicates the stability of model. In the case above is 4.204.
It is dependent of dtlongn value (60.00). This value is ok for DYNCORE_FLAG = 2 and the
model may works fine.
When integration time begins the model will print information about each timestep using
color mode:
Timestep # 1; Sim Time 60.0 [s]; Wall Time 14.875 [s]; DT 60.000 [s]; sscourn 4.204
Timestep # 2; Sim Time 120.0 [s]; Wall Time 0.983 [s]; DT 60.000 [s]; sscourn 4.204
Timestep # 3; Sim Time 180.0 [s]; Wall Time 0.947 [s]; DT 60.000 [s]; sscourn 4.204
Timestep # 4; Sim Time 240.0 [s]; Wall Time 0.998 [s]; DT 60.000 [s]; sscourn 4.204
Timestep # 5; Sim Time 300.0 [s]; Wall Time 1.061 [s]; DT 60.000 [s]; sscourn 4.204
Timestep # 6; Sim Time 360.0 [s]; Wall Time 0.906 [s]; DT 60.000 [s]; sscourn 4.204
...
If something happens during the integration it will be showed in screen. Notices in green,
warning in yellow and errors in red.
11. Contact and further info: