The time-interpolation service of GFDL's "data manager" only works correctly within single files.

The size of some datasets makes it impractical to put all time records in a single file so we often provide files for recognizable time periods like, for example, a year. This means that if the first record is, say, at 6am on Jan 1st in a file for 1979, there is not enough data in one file to interpolate correctly to, say, 3am because the previous record is another file. To overcome this limitation, we "pad" the files with the last and first records of the previous and subsequent files.

The JRA55-do data is one such dataset and this Makefile creates the padded version of the files needed to drive MOM6 in ice-ocean mode.

To generate the padded files, make sure the nco utilities are in you path. Then type:

make

To check that the generated files have the right checksum

make check

To remake the md5sums.txt checksum file:

make md5

The gnu Makefile uses ncks and ncrcat from the nco netcdf-utilities package. Use versions later than 4.2.1 to ensure the files are bitwise reproducible.

Each file found in $(DATA_DIRS) is padded by the last day at the end of the preceding year and the first day of the next year. The result is placed in $(OUT_DIR) with the .nc sufffix replaced with .padded.nc .

The JRA55-do data uses the "classic" netcdf format which does not have non-reproducible bits associated with the later formats. All this operation does is concatenate data. There are no floating point operations. This means that if we have no tool-version or date-dependent attributes added by the tools then the padded files can be bitwise reproduced which allows us to record the checksums. To achieve this, we us the "-h" command-line option for ncks and ncrcat which inhibits adding the global "history" attribute to the meta-data.

• Creating the padded files is ~1 hour using a single thread.
• Creating checksums is ~1 hour.
• Checking checksums is ~1 hour.