Custom Query (964 matches)

The switch LdefTLM used to create the tangent history NetCDF file in tl_initial.F was accidentally removed in ​src:ticket:255. Many thanks to Gregoire Broquet for reporting this bug.

Updated several algorithms:

• Replaced routines mp_gather and mp_scatter with mp_gather2d, mp_gather3d, mp_scatter2d, and mp_scatter3d. It is better to split these routines into 2D and 3D array processing to follow better the strong-typing rules in module distribute.F.
• Removed the need to provide an input tangent linear model initial conditions file (ITLname) in weak-constraint data assimilation algorithm (W4DPSAS and W4DVAR). This file is now created internally in drivers w4dpsas_ocean.h and w4dvar_ocean.h.
• Corrected bug in zetabc.F when WEST_FSRADIATION is activated. Change index kstp with know in the computation of local variable grad. Many thanks to Francesco Nencioli for ​reporting this problem.

This is a big update to various algorithms:

• Coded phase II of parallel IO. Many changes were made to the NetCDF calls in all the IO routines. All the IO is now under the umbrella of the routines defined in mod_netcdf.F.
• Corrected a parallel bug in mp_bcasti when broadcasting an integer 2D-array.
• All the broadcast routines now only have 3 arguments (ng, model, A) to avoid any compiler problem with the module interface.
• Massive changes to all the data assimilation algorithms were made to allow the adjustment of open boundary conditions (ADJUST_BOUNDARY). There is still more work needed in the error covariance convolution routines.
• Several new input NetCDF files are now needed for model, initial conditions, open boundary conditions, and surface forcing standard deviation used to convert correlations to error covariances (see s4dvar.in for details).
• Several new input/output files are now needed for model, initial conditions, open boundary conditions, and surface forcing error covariance normalization factors (see s4dvar.in for details).

For more details, please check the following forum ​post.

Many thanks to Andy Moore for his help in the data assimilation algorithms.