Detailed description is coming soon...

The lagrangian tracing of weightless particles. The scheme is based on pure interpolation (no modeling involved!). Basic idea: take the particle coordinates and predict the future location based on its current velocity. As simple as that. The scheme uses 3D velocity components and geopotential fields.

NetCDF, datetime-fortran.

First vi Makefile :

NETCDF=/opt/netcdf4-hdf5       # path to your NetCDF
DATETIME=/opt/datetime-fortran # path to your datetime-fortran libs
FC=ifort                       # your FORTRAN compiler

Normally you don't need to change other stuff.

Building:

make

Cleaning in case anything went wrong:

make clean

The source file should contain 3D velocity and geopotential fields. But it has to be properly prepared.

First, all data has to be merged into a single file. The file should have 3D variables for wind velocity components and geopotential named as u,v,w,z and coordinates (time and location).

The time dimension has to be relative (second/minutes/hours since YYYY-MM-DD HH:mm:SS) with attributes: units and calendar (normally its "gregorian").

If your data is global (reanalyzes?) then vertical velocity (w) is expected to have Pa/s unit (as reanalyzes are normally provide) and coordinates are: longitude or lon and latitude or lat for horizontal location. For vertical coordinate: levels (in Pa). All has to be vector (1D).

If you data is regional (assuming the WRF output for now, but could be anything with little modifications) then coordinates should be names as: XLONG and XLAT and expected to be 2D (moving domains are not allowed for now).

All this preparation could be accomplished by using the cdo library. For preprocessing of WRF output you might want to use prep_wrf.py script (works, but written in a very hurry).

The configuration is done using the file namelist.pt.

The domain could be regional (regional = .TRUE.) or global (periodic boundary, regional = .FALSE.). You need also set the time interval using stime and etime (this interval has to be present in the source file!).

The particles could be initiated manually (pt_grid = .FALSE.) in ASCII file named init_particles.dat (see init_particles_test_Reykjavik.dat or init_particles_test_NorthPole.dat as an example).

Or evenly spaced allover the domain (pt_grid = .TRUE.), pt_step is number of steps (grid pionts) between particles in both directions, pt_height is an elevation in meters (could be an array, but haven't tested it yet).

The accuracy could be controlled. There are two schemes provided:

1. Coarse scheme (accuracy = .FALSE.). Predicts particle location using time step from the source file. Fast but very inaccurate. Recommended only for test purposes.
2. Fine scheme (accuracy = .TRUE.). Interpolates data between source file time step. This option controlled by timestep (in minutes). The less the better but becomes very computationally heavy.

./lpt.exe src_file.nc

where src_file.nc comes from the "Data reprocessing". The output will be lptf90_src_file.nc

• Implement backward tracing
• Pix Pole issue -- sometimes particles are tend to stuck at Pole for a longer then usual (run with init_particles_test_NorthPole.dat to see)