Precision of thinned Impact Heights

[Dave Offiler]

The decision was taken for ROPP-1 v1.1 that the set of 247 Impact Heights (specified in the thinner control files 'ropp_io/data/ropp_thin_*-247.dat' should be given only to the nearest 1m.

The original values devised by Chris/Sean were to a precision of 1mm which seemed unjustified given the Level 1b specification of 6m for IP accuracy (see RTS-1030 in the URD), a 10cm limited precision in BUFR and much lower accuracy requirement for Level 2 heights (20m) for NWP applications (see RTS-1070 in URD).

It is accepted that 1m precision is quite adequate for NWP assimilation purposes and meets user requirements for this type of application.

However, Axel & Stig have pointed out that using this 1m lower precision could (theoretically) result in an additional, if small, source of error when comparing BA quality statistics prepared using the higher 1mm precision thinning control IHs, though this claim has apparently not been tested to establish the level of error induced by using different precisions. If all use the same precision, there is no problem, but for unknown reasons, Axel seem unwilling to change from the 1mm version, and DMI are using the ROPP v1.1 versions... Given that the BUFR representation is limited to 10cm precision in IP, the 1mm precision will in any case be lost to users performing comparisons using BUFR data as the source.

In view of the 6m nominal IP accuracy, BUFR limitation and potential for induced errors, we suggest a compromise precision of 10cm for the IHs in the thinning control files, which is the best that BUFR can provide (unless the template is upgraded; not a short-term option)

The impact of using 10cm and 1m vs the reference 1mm precision on the (O-B) statistics using a common set of RO & NWP profiles should be established before the 10cm control files are included in the ROPP release (nominally v2.0, possibly as a post-v1.2 optional update). A user can point the ROPP tools (ropp2ropp, ropp2bufr) at any control file they wish from the command line. If 10cm files are prepared, Stig will perform such a test (TBC)

Whether or not 10cm precision is agreed for ROPP releases, a common precision needs to be agreed and adopted by EUM and the SAF for (non-BUFR) validation. Note that end users (including MetO) will be using BUFR products, so will be limited to a 10cm precision even if 1mm (or 10mm) were in the thinner control files).

[(none)]

Milestone 2.0 deleted

[Huw Lewis]

Response from Axel (24/09/08)

The main impact of the precision in the old comparisons was in the multi path region, here it could lead to large differences. With the sorting now being implemented, we are less susceptible to it. The test data set I provided is anyway only in BUFR and PFS, so what we actually use in house as thinned levels is not really visible up to it's full precision once the data is written to BUFR, as you said. Thus I also expect to see very little differences in the sorted EUMETSAT based thinning and the ROPP one, even if the used levels do not agree up to the last cm. 

In the end it is up to you what exact levels you are using as long as they show the consistency we expect. We for now will not change the operationally used levels, any change in the operational processing is a major effort and we have quite a few more important tasks on our plate. But in the long term we will look at it again when implementing the thinner in the PPF.

Cheers,

           Axel

DMI request at GRAS SAF PT 4 to use 'full resolution' EUM thinner levels in v2.0. This file is therefore included in the ROPP-2 distribution. Testing by Stig showed small (but non-negligible) impact of using 0.1 m resolution thinned levels compared with EUMETSAT levels. See attached files.

The other thinner files provided with ROPP-2 are however now provided to the nearest 10 cm (see [1782]).

Ticket closed as fixed - ROPP thinner files are given to nearest 10 cm (BUFR resolution), DMI use 'local' ropp_thin_eum-247.dat file distributed as part of ROPP package.

[Huw Lewis]

Comparison of DMI thinned BAs with EUMETSAT using 1 cm thinner control file (same as at EUMETSAT)

[Huw Lewis]

Comparison of DMI thinned BAs with EUMETSAT using 10 cm thinner control file