Custom Query (964 matches)

It turns out that we need to follow the point association rules for array IJwater in the call to mp_scatter in input routines nf_fread2d.F, nf_fread3d.F, and nf_fread4d.F when the READ_WATER option is activated. Although these input routines are not inside a module, the mp_scatter routine is and this array is passed as a pointer using the SCALARS(ng)%IJwater synthax. This problem is also found in routine white_noise.F.

This is similar to ​src:ticket:228 and discovered using ifort 10.1. Many thanks to Mathieu Dutour for reporting this problem.

A memory leak was found in nf_fread2d.F when the option READ_WATER was activated. To avoid such leaks in the future and facilitate debugging nf_fread2d.F, nf_fread3d.F, and nf_fread4d.F were put inside modules. Therefore, all the routines that use these input routines were changed to have the correct pointer association in the arguments. Currently, all the generic NetCDF interface is inside modules.

Corrected bugs in wpoints.F for serial and shared-memory applications.

Eliminated a potential problem in extract_obs.F and ad_extract_obs.F when the observation depth is given in meters (negative value) instead of fractional model level (positive value). Observations shallower or deeper than the model associated variable, will be flagged out and ignored. Users should be carefull when preparing the 4DVar observation NetCDF file. If assimilating satellite SST, assign as a depth the model top level (N(ng), positive value) instead of zero or negative depth values. The depth value provided to altimetry SSH data is irrelevant but it is recommended to use a zero value.

Updated obs_read.F and obs_write.F to use new generic NetCDF interface (read and write routines in mod_netcdf.F). Also corrected couple of bugs and cleaned obsolete code and variables.