From:	Stig Syndergaard [ssy@dmi.dk]
Sent:	10 February 2015 12:06
To:	Burrows, Chris
Cc:	Culverwell, Ian
Subject:	Re: ROPP_PP

Hi Chris,

I agree with your observations for the lat, lon, azimuth interpolation in the code, and from 
your figures it looks like you've found a good solution. Why only temporary?

No doubt that the occ_tool (and the invert_tool) has some issues. Maybe we should 
make a coordinated effort to get them solved for the next ROPP release. We have the 
ones from Kjartan, my Struct.Uncert. report, our DMI tickets 
(https://trac.romsaf.org/ropp/report/10) and now your investigations with GNOS - 
probably with some overlap.

I also think we should aim at merging the occ_tool and the invert_tool (I think I've 
mentioned that earlier), it would make maintenance easier.

Ian: Would it be possible to schedule all this for ROPP 9?

-Stig

On 2015-02-05 14:38, Burrows, Chris wrote:
> Many thanks Stig,
>
> That report is very helpful indeed - it's good to know that you have seen the same 
features (e.g. Fig 2.5).  I've come up with a temporary fix for the lat/lons (see attached), 
though it's not a robust solution.
>
> Please don't let this distract you from your other work, but any input would be gladly 
received!
>
> Thanks again
>
> Chris
>
>
>
> -----Original Message-----
> From: Stig Syndergaard [mailto:ssy@dmi.dk]
> Sent: 05 February 2015 13:18
> To: Burrows, Chris
> Cc: Culverwell, Ian
> Subject: Re: ROPP_PP
>
> Hi Chris, Ian,
>
> I'll have a look at this over the next couple of days (into next week); today I am just 
swamped in other work. I remember from the SU project some years ago that the lat, lon 
vector was a bit odd. And also a kink at 25 km in std.dev. I've attached the report that I 
did back then. At the end (section 3) there is a summary of the issues that I found. It's 
been on my mind to fix these things in ROPP for a while, but I never seem to get around 
to do it (always something more urgent).
>
> Best,
> -Stig
>
> On 2015-02-05 13:19, Burrows, Chris wrote:
>> Hi Stig, Ian,
>>
>> Just a quick update;
>>
>> It seems that the IPs are calculated for GO, and then, for the WO processing, the IPs 
below 25km are overwritten with uniformly spaced levels (keeping the size of the array 
the same). The lats, lons and azimuths *aren't* mapped onto these levels, so the final 
interpolation 
(https://trac.grassaf.org/ropp/browser/ropp_src/trunk/ropp_pp/tools/ropp_pp_occ_tool.f90
#L731 ) is mismatching lat,lon,azim with the post-WO impact parameters. Maybe we 
should store the pre-WO values for this interpolation step. It's not a huge deal, but would 
be important for TPD in assimilation, and 2D operators.
>>
>> As for the jump in standard deviations, I wonder if it's related to the fact that the WO 
processing outputs IPs on a fine grid (~5m in the case I was looking at). This is then 
interpolated onto a coarse grid, so I fear that a lot of information is being lost (this is 
probably consistent with large noise but small correlations).
>>
>> I'll keep looking into this, but I'm sure I'm not qualified to mess around with the WO 
settings - scary stuff...
>>
>> Thanks
>>
>> Chris
>>
>>
>>
>> -----Original Message-----
>> From: Burrows, Chris
>> Sent: 04 February 2015 18:28
>> To: Stig Syndergaard
>> Cc: Culverwell, Ian
>> Subject: ROPP_PP
>>
>> Hi Stig,
>>
>> I was looking at some sample data from the Chinese GNOS instrument with Ian today 
- they are using ROPP for the preprocessing. It looks pretty good but there is a sharp 
jump in the BA O-B standard deviations at 25km (see attached O-B stats and vertical 
correlations).  This presumably is related to the WO/GO transition (Kjartan made 
reference to something similar here: https://trac.grassaf.org/ropp/ticket/293).
>>
>> We did a bit of further digging and what we found may or may not be related, but it is 
certainly confusing....
>>
>> In the output file, the impact parameters are spaced equally, but there are sudden 
jumps in the tangent point lat-lons corresponding to IH=25km. The values do correspond 
to a sensible track; the transition is just due to a change in the data sampling. The 
azimuth also shows a suspicious jump. These are plotted in the attached 'coords.png' 
file (note that the jagged steps in the azimuth are due to BUFR precision; they are not 
present in the full-resolution files, though the 25km feature is still there).
>>
>> The bottom right plot is the 'angular distance', sqrt((delta lat)*2)+(delta lon)^2)) as a 
function of impact height (i.e. a vertical cross section of the TP trajectory) - this looks very 
suspicious!
>>
>> Similar plots are produced with GRAS data using ROPP, so we suspect there may be 
a bug. Do you have any thoughts about this? Perhaps the lat/lons are being overwritten 
incorrectly for the WO part of the processing (still sampled against time rather than a 
fixed IP grid?).
>>
>> Thanks in advance for any comments, (please copy Ian into a reply 
>> too!)
>>
>> Chris
>>
>>
>> ---
>> Chris Burrows   Satellite Applications Scientist
>> Met Office   FitzRoy Road   Exeter   EX1 3PB   United Kingdom
>> Tel: +44 (0)1392 88 6449
>> E-mail: chris.burrows@metoffice.gov.uk   http://www.metoffice.gov.uk
>>
>>
>

--
_________________________________________
Stig Syndergaard, PhD
Danish Meteorological Institute
Lyngbyvej 100
DK-2100 Copenhagen

Phone: +45 39 157 408
E-mail: ssy@dmi.dk
_________________________________________
