Few updates

[arango]

• Changed the time clock function to measure elapsed time in shared-memory parallel applications from cpu_time to omp_get_wtime. It turns out that cpu_time returns the cumulative elapsed time of all parallel threads instead of a single parallel threads as omp_get_wtime. This resulted into inflated elapsed time values during profiling. Many thanks to Simon Spagnol for ​bringing this to our attention.

• Changed calls to create and open NetCDF files when parallel I/O is activated using CPP options PARALLEL_IO and HDF5. We are using now nf90_create and nf90_open with F90 optional arguments in mod_netcdf.F instead of nf90_create_par and nf90_open_par, respectively.

• The ​release 4.1.2 of the NetCDF library made several internal improvements to parallel I/O performance. However, it broke the logic of writing of scalar variables to the NetCDF-4 (HDF5 format) file. We get now an error when writing scalar variables. If parallel I/O is activated, we need to use an older version of the library until we solve this problem. We don't know if this is a new bug in the library or we need to create/write scalar variables differently. Unfortunately, there is no documentation for how to do this.

• Corrected a bug in the week constraint 4D-Var (W4DPSAS and W4DVAR) when model error is included. It turns out that we were not clearing the model error forcing arrays at the start of each inner-loop in routine frc_weak.F. Many thanks to Andy and Polly for discovering and fixing this serious bug.

• Modify inp_par.F to allow different values of NHIS and NADJ in week constraint 4D-Var applications. Sometimes, it is necessary to have NADJ<<NHIS. Many thanks to Andy for suggesting this change.