Summary of meetings

on January 10 and February 21, 2008

 and progress during March

 

 

Materials are posted at

 

http://www.emc.ncep.noaa.gov/research/JointOSSEs/NR/Jan-Mar08/



Joint OSSE meeting January 10, 2008

 

Attendees:

Hans Huang, Lars Peter, Riishojgaard, Eric Bayler,  Oreste Reale, Jack Woollen, Haibing Sun, Gary Wick, Yuchang Song, Tong Zhu, Chaojiao Sun, Juan Jusem, Yoshiaki Sato, David Groff, Michiko Masutani

 

Teleconferees:

Erik Andersson - ECMWF

Tom Schlatter, Yuanfu Xie, Nikki Prive, Dezso Devenyi - ESRL

Chris Hill, Valentine Anantharaj - GRI/MSS

 

 

Joint OSSE meeting February 21, 2008

 

Attendees:

Oreste Reale, Jiayu  Zhou, Jack Woollen, Haibing Sun, Daryl Kleist, Michiko Masutani

 

Teleconferees:

Erik Andersson - ECMWF

Yuanfu Xie, Nikki Prive, Dezso Devenyi - ESRL

Chris Hill - GRI/MSS

 

Out with Flu:  Tom Schulatter, Tong Zhu

On Travel:  Fuzhong Weng

Conflict with another meeting:  Harper Pryor, Runhua Yang, Ron Errico (sent a summary report)

Information about DWL data provided by Ad Stoffelen.

 

Materials posted at

http://www.emc.ncep.noaa.gov/research/JointOSSEs/NR/Jan-Mar08/

Meso Scale OSSE for MTGIRS by Hans Huang (NCAR)

     200801EMC-MTGIRS-OSSE.pdf 
     MTGIRS_MR_fin.pdf    


Precursor Run: Yuanfu Xie

     YXie_Report_080110.ppt 
Agenda and summary by Michiko Masutani
   including summary of discussion on meso-scale OSSE

            Agenda.080221.ppt

Tropical cyclone diagnostics by Oreste Reale

            Reale_T799_October_Nature_Run.ppt

Arctic in Nature run   by Nikki Prive

            Prive_arctic_ice_080221.ppt

Update in precursor run  by Yuanfu Xie

            XieY_Report_080221.ppt

 

           

1.  Hans presented on the mesoscale regional OSSE for MTGIRS.   

 

MTGIRS is similar to GOESR.  The OSSE is conducted for a quick look.  The NR is only one week long and was produced using MM5 with 4km resolution, and data from the WRF model was used for assimilation.  Radiance retrieval data are generated by EUCOS from simulated radiances.  The first set of retrieval data contains errors that are too large.  Due to the correlated error in the improved retrieval data, 100km sampling was required.  He showed that radiance has to be assimilated in order to achieve the use of high resolution data.

 

Many are impressed by the speed and detailed work done but are also convinced that there are no reliable results.  Although he made clear the limitations and problems, Tom Schlatter found that Hans drew too many conclusions that he can’t afford.  Fuzhong is afraid that managements will tend to believe that theOSSE for GOESR is done. 

 

 

2.  Negative moisture in Antarctic.

 

Yuanfu presented strange results from the precursor run.  (1/10 and 2/21 meetings)

 

He found that after 20 days the RMSE from the NR and analysis with simulated conventional data is as large as RMSE from the NR and free run.

 

Jack and Michiko joined the investigation.  We suspected unreasonably bad data but removing the bad data or using variational QC did not help.  Increasing the resolution to T126 seems to resolve the problem.  Testing with a smaller time step did not help.  Michiko confirms that many data are just rejected.

 

Daryl’s comments (2/20/2008 meeting)

"It looks like there is a pretty bad drift in the temperatures over Antarctica.  I wonder if the drift is the result of a model problem, and that there is no data (i.e. satellite data) to control and correct the problem.  Just look at the surface temperatures over this region from the 10th to the 20th, there is a pretty big drift downward.

 

The regions where I see the negative moisture problem develop also correspond to bullseye's in really cold skin temperatures that develop (from the surface flux files). "

 

Jack suspected bad data but it seem to be data are all right.  Setting factqmin/max=0.0 or increasing the resolution to T126 helped.

 

Discussion on this subject is posted at

http://www.emc.ncep.noaa.gov/research/JointOSSEs/forums/Precursor_run/

It is not linked from Joint OSSE web site but there is some additional discussion at

http://www.emc.ncep.noaa.gov/research/JointOSSEs/forums

 

This problem seems to be caused by a combination of moisture corrections in the GSI and GFS model.  John Derber is working on this problem.

 

In order to coordinate with the GSI developers, it is desirable to upgrade our OSSE system.  Since Diane Stokes is working on a new script, we will wait for her work to be finished.  For the time being, we will continue work using the March 2007 version of GSI.

 

 

3.  Help line and FAQ site for Joint OSSE.

 

Michiko started a note to record help requests.  Any small help provided will be recorded so help given can be acknowledged appropriately.  Michiko, Yuanfu, Yucheng and Tong will form a help line to answer basic questions and direct users to the right people, and build a knowledge base for OSSE.  Please direct OSSE questions to these people.  NCEP operational staffs are not able to answer the same questions many times over again.

 

Michiko will also work on an FAQ web site.

            Question 1:

            What is the difference between gfs and fnl in GDAS scripts?

The difference between 'gdas' and 'gfs' is actually in the analysis.  In operations, we create 2 analyses per cycle.  The first analysis is run shortly after the synoptic time (I believe that's 1.5 hours right now), and is used as the analysis for our operational, 15-day GFS forecast.  However, a later analysis is run at 3 hours past the synoptic time (when more data has trickled in), where a shorter (GDAS) 9-hr forecast is run.  This 9-hr forecast is presumably of slightly better quality, as its analysis contains more observational information.  This 9-hr GDAS forecast is then used as the subsequent background for the next cycle's GFS AND GDAS analyses.

 

So, for any given cycle (00, 06, 12, & 18) we have two analyses (GFS and GDAS), as well as two sets of forecasts (GFS, out to 15 days, and GDAS, out to 9 hours) from their respective analyses. (Daryl Kleist)

 

Question 2:

What is the plan for incorporating RTTOVS into CRTM?

Originally, an option to select either OPTRAN, RTTOVS, or SARTA in the CRTM was planned.  However, the plan seems to have changed.  OPTRAN, RTTOVS, SARTA and the other RTMs each has its own strength and weakness, so CRTM teams are working on combining good parts of each RTM model into the CRTM.  They are hoping to have an experimental version in eight month’s time but it will take another half year or so to become operational.  Their goal is to improve the CRTM  for operational data assimilation.

 

We thought that if there is an option in CRTM, we can take advantage of it and use different RTMs for simulation and assimilation.  It seems that if you wish to use RTTOVS or SARTA to simulate observational error you may have to use the existing RTTOVS and SARTA without CRTM.

(David Groff and Michiko Masutani communication with Paul Val Dest and Yong Han)



4. Progress in simulation of radiance at GMAO.

 

Ron Errico (1/9/2008 meeting)

We are proceeding with simulating the radiance observations. It took us a while to get all the correct CRTM components. The initial tests of our algorithm for simply including (and later tuning) cloud effects appear to be good. We are planning on producing both IR and MW radiances. We will make the BUFR datasets and software available when they’re complete.

 

Ron Errico (2/20/2008 meeting)

I have been sick with the flu since last week.  Also, Runhua and I have 2 meetings Thursday afternoon, so neither of us will be attending the OSSE meeting this month.  Our status report follows:

 

1. The software development for our first attempt at a GSI DA with all currently used observations is nearly complete.  We are still working on some final pieces for the IR radiances and expect to have it working for both AIRS and HIRS by the middle of March.  The code has been developed with the hope of also applying it to AMSU with minimal changes, but since we as yet know little about that instrument, we can't yet say what special effort may be required. All the simulated observations have some random error added and the datasets are in bufr format.  They are all produced directly from the sigma-level data on the reduced Gaussian grid.

 

2. The new version of the GMAO GSI adjoint is now working, so all our tuning of the observation simulations will be conducted using that tool, rather than running OSEs.  If you want to learn about this tool, Ron Gelaro will present a seminar describing its use and validation on Monday Feb 25th at 2:30 at GSFC.

 

3. In order to save processing time, we are producing radiances for only a thinned set of locations.  The thinning is not as drastic as the GSI applies so it still allows the GSI to apply its own data selection algorithm, but to a reduced set. Also, we plan to produce AMSU results only over oceans, since the current version of GSI that we use does not use such obs over land. This makes our task much easier, with both fewer issues to consider and less data to process.

 

4. We hope to have tunned results for all observations by this summer, which allows us several iterations of 2-4 week assimilation experiments to conduct before then.  These will be for a winter period.  The tunning concerns parameters used to specify cloud effects and error statistics. Once we have the tunned results, we will make available the bufr datasets we have and the software we used to produce them. We will also then begin to complete the full year of simulated observations.

 

5. Runhua is currently busy with some other projects, but will return to the OSSE work shortly. I have other things I am supposed to be doing now, but have instead been working on the OSSE code development, augmenting what Runhua has done. Some components took us a long time to develop; if someone who understood some critical pieces that we had to get from elsewhere, we could have saved 6 weeks at least. None-the-less, I like what we have thus far.  It seems easy to both use and modify, being in the form of modules independent of the massive GSI code.

 

6. After this first set of tuned simulated obs are produced (this summer) I will work on further improvement of the realism of the observations.

 

Although succinct, this report is fairly complete and accurate.

 

Ronald Errico

 

Note:

Tong provided his scripts to GMAO and spent some time to explain the scripts.  Jack produced 91 level data for HIRS to speed up work at GMAO.  Haibing provided his code to Emily Liu.  Emily spending a lot of time helping with understanding the CRTM.  Ravi Govindaraju is working as a BUFR expert at GMAO for this project.

 

Follow up message from Ron Errico:

 

We are producing thinned observations, but done so in a way that the GSI can still do its own additional data selection as it is designed to do. So, for example, we don't create obs for all 300,000 AIRS obs available for one 6hr period.

 

It takes about 1 day of single-processor computer time to create all simulated conventional obs for 1 year. To create all satellite obs for 1 day using a single processor will take about 2-3 hours using my current degree of thinning. These are obs both with and without errors.

 

We will not release our software or data until we have tested and tuned it with respect corresponding data impacts in the OSSE and real analysis. Otherwise, people will use erroneos data and either complain or simply ignore the bad results.

By the way, although the OSSE period includes the winter of 2005-2006, it appears that at least Feb of that period is very anomalous regarding forecast skill in the real case. So, it may be difficult for the OSSE simulated skill metrics to match the real case during that period.

 

 

5.  Format of DWL data.

 

Michiko (2/19/2008)
David Tan gave me a ppt presentation and L2BP_Release1.31 last year.  Does that apply to the format you will use when simulating DWL data?

http://www.emc.ncep.noaa.gov/research/JointOSSEs/NR/Feb08/Talk_Tan_L2BP_CLRC20070608.ppt
Michiko

 

Ad Stoffelen (2/20/08 )

No, this is not our intention. Our DWL simulator, called LIPAS, uses ECMWF inputs and takes account of subgrid variability statistically (cloud, aerosol). This module has not been developed for the E2S (as described by Tan et al.), but is statistically compatible. LIPAS delivers BUFR files as these were delivered before to the ECMWF OSSE data base.  This is what we intend to provide.
 
Michiko (2/20/08 )
I am wondering the format we will get from the real ADM.  If we set up a data assimilation system for LIPAS output, is it easy to switch to real ADM data.  Is the format discussed by Tan the format we will receive from the real ADM?

Ad Stoffelen (2/20/2008 )
There is no connection between the EE format discussed by Tan et al. and the LIPAS BUFR. We are currently not in a position to provide you with Aeolus data as you will get them in NRT, but we will be able closer to the launch. In principle we could deliver:
- L1B data and a L2B processor, standalone and portable source code
- L2B data (LOS wind profiles, optically classified)
The L2B processor needs NRT T meteo input with accuracy better than 10K.
T sensitivities and references are delivered in the L2B data, so any Met centre may correct for his own best T, but there appears not much reason to do so (all are better than 10K). The T (and less p) sensitivity is important for the width of the molecular Rayleigh distribution, which is used in the processing.
 
Both datasets may be delivered in ESA EE format and BUFR. The former is about ten times the size of BUFR. What would you require in terms of format and timeliness?
 
Please ask for further clarification if need be. More details are available.
 
Michiko (2/20/08 )
Oliver said we have to wait before starting work on L2b L2c data, until the software becomes more finalized.  I will keep your message as a record and we will keep L2b L2c data in our minds. 
 
Stoffelen (2/20/08 )
Quite right.
 
Further discussion will be posted from discussion forum.
 

 

6.  Presentation at AMS meeting and LWG(Lidar working group)  meeting.

 

Michiko presented progress in the Joint OSSE at IOAS-AOLS at the AMS meeting and LWG meeting . Presentation made at LWG is:

http://www.emc.ncep.noaa.gov/research/JointOSSEs/NR/Feb08/Joint_OSSE_LWG_Feb08.v2.ppt

 

Presentation by Dave Emmitt at the AMS meeting is posted at

http://www.emc.ncep.noaa.gov/research/JointOSSEs/NR/Feb08/Emmitt_Potential_Impacts_of_a_Space-based_Doppler_Wind_Lidar.ppt

 

 

7.  Data distribution.

 

NASA portal maintenance was moved from SIVO to NCCS Data Portal.

NCCS = NASA Center for Computation Sciences
The NCCS is managed by CISTO.
 CISTO = Computational and Information Sciences and Technology Office
Contact person Harper Pryor  Harper.Pryor@NASA.gov .


Harper Pryor is the Programs Development Manager for CISTO and the SMD User Advocate to the NCCS.

Alindo is working on reorganizing the data.

A full resolution T511 NR has been posted.  The data is organized for each forecast hour and support access is by gradsdod.  See details in the attached note.

 

Verification data was delivered to NRL/Monterey, University of Utah, and MSU.

           

Verification data:  1x1 deg and 0.5x0.5 deg lat lon gridded data for pressure, potential temperature and surface variables.  Reorganized for each time period and grads control files have been created.  Links to make the ctl file work are also provided.

 

 

8.  Tropical Cyclone in T799 NR.

 

Oreste showed that Atlantic tropical cyclones in the T799 NR have some unrealistic    features.

·        AEJ is 40% weaker than climatology;

·        Atlantic TC activity contains some highly suspicious tracks;

·        Eastern Pacific seems to present excessive proliferation of weak TCs;

·        The intensity of the strongest ATL systems is not superior to T511;

·        Different behavior in different basins;

·        Structure of some intense systems not very satisfactory in terms of scale and size of eye-like features.

 

His preliminary conclusions are

·        The increased resolution does not necessarily provide stronger confidence in a much better Nature Run for the tropics;

·        Representation of TC activity does not immediately appear superior to the T511;

·        Caution should perhaps be used in adopting this NR for applications centered on future instruments targeting hurricanes;

·        Further investigation is needed.

 

Erik is putting together the response from ECMWF.

 

9.  Arctic in NR. 

 

Nikki Prive showed that

         Stratospheric ‘cold pole’ - without input from obs, model develops abnormally cold polar region in winter with strong polar vortex. Common problem with global models.

         Surface temperatures are strangely high over the arctic.

        Disconnect with temperatures above surface layer?

         Low level inversion too weak - impacted by many processes.

         What is the energy balance over ice? SHF/LHF/longwave/cloud impacts?

         Sea ice and SST are applied from 2005/6 fields, not interactive with model atmosphere.

 

The reference article:

Beesley, J., C. Bretherton, C. Jakob, E. Andreas, J. Intrieri, and T. Uttal (2000), A comparison of cloud and boundary layer variables in the
ECMWF forecast model with observations at Surface Heat Budget of the
Arctic Ocean (SHEBA) ice camp, J. Geophys. Res., 105(D10), 12337-12349.

http://www.emc.ncep.noaa.gov/research/JointOSSEs/references/Beesley_ECMWF_SHEBA_Arctic_2000JD900079.pdf

 

 

10.  Upgrading GSI script.

 

There have been  numerous changes in the GSI code and scripts since we copied them in March 2007.  Diane Stokes is working on putting together a stand-alone subdirectory for the parallel scripts.  Probably it is better to wait for her work to be completed before making the next upgrade.  Moisture constraint procedures are still being developed.

 

 

11.  Data distribution.

 

Verification data have been sent to NRL/Monterey, University of Utah and Mississippi State University (MSU).  The complete data have been posted on the NASA portal.

 

All model resolution data have been posted to the NASA portal.  T799 NR 0.5x0.5 data have also been posted.  T511 1x1 data have been posted from NCAR.

 

When MSU downloaded model files it took about 30-40min to download 1.7GB data, which is the 1hrly out put for the T799 NR.  There are 840 files for each T799 NR.  Therefore, it will take about 500 hours (21 days) to download each T799 NR. 

 

Downloading a 2GB file from Japan took 5 hours.

 

 

12. Japanese effort to LETKF.

 

Takemasa has downloaded all conventional data generated by Jack.  He had some trouble in reading the bufr data, but it seems to have been resolved.  Takameas Miyoshi and Takeshi Enomoto are trying to assimilate these data using LETKF at Earth Simulator.

 

 

13.  OSSE for COSMIC data.

 

 Lidia Cucurull is working on a proposal.

 

14.  Utah.

 

Zhaoxia Pu has received simulated conventional data and verification data.  She is working on a proposal for a regional OSSE to evaluate DWL.

 

15.  Mesoscale OSSE at MSU.

 

Chris Hill will produce a summary of the meeting on April 8.