Pre-sort impact parameters before thinning

[Huw Lewis]

In order to resolve small discrepancies between EUMETSAT and GRAS SAF implementations of the ROPP thinner, a pre-sort of impact parameters has been agreed. This needs to be implemented in ROPP based on code changes suggested by Axel [1760].

[Huw Lewis]

Input data file with full resolution bending angle data from PPF

[Huw Lewis]

Output file with thinned bending angles generated at EUMETSAT

[Huw Lewis]

Output file with thinned bending angles generated using ROPP v2.0 thinner

[Huw Lewis]

Plot showing consistency between EUMETSAT and ROPP thinned bending angles (test_e.nc vs test_t.nc)

[Huw Lewis]

Plot of errors in thinned BA using ropp_thin_log-247.dat rather than ropp_thin_eum-247.dat file

[Huw Lewis]

The ROPP thinner code for v2.0 has been modified in line with Axel's suggestions ([1760]). See [1767] for details.

These updates have been made in a more consistent way with the rest of ROPP than in Axel's changed code, so there is no 'EUMETSAT' thinner option as such. It will be possible to run using the EUMETSAT thinner file however, to get consistent impact heights, but the thinning method is still just using log interpolation.

At the moment, sorting is now the default for all thinning methods - this makes sense if we consider the algorithm is 'designed' for sorted data. A 'I don't want sorting' type flag may be added at a later stage, but this is not a priority.

I have also included a suggestion from Axel (implemented in a more efficient way than Axel suggested) that we always output the requested number of thinned levels, even if there are insufficient (or no) input data. This makes concatenating data files together a lot easier. This is related to this outstanding ticket (#120) although more thought and work may be required to fully answer this (i.e. 'short' profiles are now set to missing, but put in an output file at the requested resolution). I have tested this code with some data files Axel provided (I concatenated the first 2 files in the tar file I pointed Stig to in an earlier email). These have been attached to this ticket. Also attached are the comparison between thinned bending angles generated using ROPP and the thinned bending angles from EUMETSAT. When using the same thinned impact heights file (ropp_thin_eum-247.dat) the output is identical. I also have a plot showing errors of up to 0.5% when using the 'ROPP' levels for this method in file ropp_thin_log-247.dat.

Ticket remains open for further iterations and testing by GRAS SAF at DMI.

[Huw Lewis]

ropp_io distribution for testing at DMI

[Huw Lewis]

EUMETSAT thinner control file

[(none)]

Milestone 2.0 deleted

[Huw Lewis]

DMI to implement ROPP v2.0 thinner in operations.

Ticket closed as fixed.