Opened 4 years ago

Last modified 3 years ago

#184 assigned

Preliminary investigation of all-sky microwave radiance assimilation

Reported by: Jin Lee Owned by: Jin Lee
Priority: major Component: ACCESS model
Keywords: all-sky microwave observation assimilation Cc: Peter Steinle, Fabrizio Baordo, Chris Tingwell


ECMWF implemented its first all-sky microwave radiance assimilation in 2009 and since then it has been steadily improving its 4DVAR's ability to assimilate cloud-affected microwave radiances. This resulted in substantial gains in their forecasts. The all-sky methodology as described in Bauer et al. (2010, QJ) provides a scientifically sound and consistent approach to the assimilation of cloudy radiances and has a great scope for further development.

The all-sky methodology at ECMWF was developed over many years and it involved significant changes to IFS. The effort required to implement it in our NWP system is beyond our current ESM resources. However if we can draw on the technical and scientific capacity of UKMO through collaborative partnership it may just be possible.

But first, here I will try to investigate whether or not it's possible to apply the all-sky methodology to our global NWP system. I will follow Geer and Bauer (Tech. Memo. 620, ECMWF) very closely and test whether or not there is any inherent technical and/or scientific limitation in our OPS, VAR and UM which prevents the implementation of all-sky approach.

Change History (7)

comment:1 Changed 4 years ago by Jin Lee

comment:2 Changed 4 years ago by Jin Lee

Set up a suite that assimilates SSMIS

  1. au-aa300 (Copy au-aa094)
  2. Include SSMIS in suite.rc by comparing with au-aa099
  3. Luke-warm start 20141119 18Z
    1. iau-start-dump from 20141119 12Z
    2. an_inc from 20141119 18Z
    3. an_smc from 20141119 18Z
    4. from 20141119 18Z
  4. First full assimilation and forecast cycle is 20141120 00Z

RUN 1: Spin-up for bias correction

  • Start date: 2014112000 (luke-warm start; first full OPS, VAR and UM were for the cycle time of 2014112000)
  • End date: 2014120318
  • Stationlist: bom/common/aps2_r00.v381
  • Bias coeffs: F17 and F18 are in Bufr (F16 died in 2013); Vinod initially used F16 coeffs for F17 and F18 (based on bom/common/aps2.2014022718?)
  • SatRad coeffs:
    • SatRad channel selection ( and - some channels switched off
    • SatRad control file ( - to include writefile output spec

Note that above plan was not followed through but instead I followed the plan outlined below

Last edited 4 years ago by Jin Lee (previous) (diff)

comment:3 Changed 4 years ago by Jin Lee

Revised plan

After talking to Fabrizio the first thing to do is to assess and identify the short-comings of the forward model (RTTOVS) used in our current assimilation in cloudy conditions. Because radiative transfer calculations for window channels is simplest (I think?) I'll first focus on the window channels of SSMIS:

Ch Centre Freq (GHz)
12 19.35 (H)
13 19.35 (V)
14 22.235 (V)
15 37 (H)
16 37 (V)

So the first task is to generate o-b for SSMIS (possibly also for AMSU-A2 and MHS window channels) using the control APS2 model background. This means SSMIS has no effect on the analysis - and consequently no effect on model background. OPS task is used merely to generate statistics.

The experimental suite would do:

  1. get_bufr for ATMS, ATOVS and SSMIS
  2. convert Bufr to ODB for ATMS, ATOVS and SSMIS
  3. copy previous cycle's cx background and current cycle's bgerr from the control, au-aa094
  4. run Ops_atms, Ops_atovs and Ops_ssmis
  5. archive SatRad writefile for ATMS, ATOVS and SSMIS
  6. archive Bstats for ATMS, ATOVS and SSMIS

Plan of runs to do:

  1. Spin-up for bias correction: 2014112000 - 2014120318
  2. Update bias regression coefficients
  3. Run after first update of bias regression coefficients: 2014120400 - 2015010418
  4. Update bias regression coefficients
  5. Run after second update of bias regression coefficients: 2015010500 - 2015020418
Last edited 4 years ago by Jin Lee (previous) (diff)

comment:4 Changed 4 years ago by Jin Lee

OPS control files

au-aa300 was derived from au-aa094.

In this experiment I will be running only Ops_atovs (for o-b from AMSU/MHS) and Ops_ssmis. So only those OPS control files to do with ATOVS and SSMIS are needed but they need to be changed:

  • RTTOV7_coeffs
    • no change from au-aa094; later when RTTOV-SCATT is used these will need to be changed drastically
    • Q. v34 is used - is this suitable?
  • SatRad_biases
    • start from available files; since bias regression coefficients will be updated I expect no significant effect
  • SatRad_coeffs
Bmatrix Rmatrix Rmatrix_Var ThinControl bias.qc channels channels_Var control
SSMIS au-aa099 au-aa099 au-aa099 au-aa099 au-aa099 au-aa099 au-aa099 au-aa099 (+writefile)
ATOVS au-aa094 au-aa094 au-aa094 au-aa094 au-aa094 au-aa094 au-aa094 au-aa094 (+writefile)
ATMS au-aa094 au-aa094 au-aa094 au-aa094 au-aa094 au-aa094 au-aa094 au-aa094 (+writefile)
  1. for SSMIS use the version from au-aa099 + SatRad writefile generation
  2. for ATOVS use the version from au-aa094 + SatRad writefile generation
  3. for ATMS use the version from au-aa094 + SatRad writefile generation
  • StationLists
    • use au-aa094 but add F17 and F18; note UKMO didn't change SSMIS thinning distance when PFM resolution increased from N108/N216 to N144/N320
Last edited 4 years ago by Jin Lee (previous) (diff)

comment:5 Changed 4 years ago by Jin Lee

Status: newassigned

comment:6 Changed 4 years ago by Jin Lee

Run 1: Spin-up for bias correction

Period: 2014112000 - 2014120318

Cycle time Failed task Reason for failure Action taken
2014120100 - 2014120112 Ops_ssmis SSMIS Bufr file zero size reset task to succeeded
2014120500 Ops_atovs ATOVS Bufr contained obs from ATOVS/Aqua (EOS 2, 784) but bias coeff file didn't have coeff for ATOVS/Aqua Manually added bias coeffs for ATOVS/Aqua from ukmo/v12
PostArchive fails to copy Bstat files to 'mdss' directory PostArchive task has the call sequence, arc2sam.ksh -> arc2sam.ksh -> arc2mdss.ksh. In arc2mdss.ksh there is a line,

if [[ -d $stats_dir ]]; then

Because $AN_DIR doesn't exist the 'if' was not entered
Decided not to fix the problem but instead copy manually Bstat files using,

Last edited 4 years ago by Jin Lee (previous) (diff)

comment:7 Changed 3 years ago by Jin Lee

Cc: Peter Steinle Chris Tingwell added
Note: See TracTickets for help on using tickets.