1 / 26

Assimilation of Satellite Radiances into LM with 1D-Var and Nudging

Assimilation of Satellite Radiances into LM with 1D-Var and Nudging. Reinhold, Christoph, Francesca, Blazej, Piotr, Iulia, Michael, Vadim DWD, ARPA-SIM, IMGW, NMA, RHM COSMO General Meeting, Cracow 15-19 September 2008. - plenary session -. 1DVAR + Nudging = Nudgevar.

damali
Download Presentation

Assimilation of Satellite Radiances into LM with 1D-Var and Nudging

An Image/Link below is provided (as is) to download presentation Download Policy: Content on the Website is provided to you AS IS for your information and personal use and may not be sold / licensed / shared on other websites without getting consent from its author. Content is provided to you AS IS for your information and personal use only. Download presentation by click this link. While downloading, if for some reason you are not able to download a presentation, the publisher may have deleted the file from their server. During download, if you can't get a presentation, the file might be deleted by the publisher.

E N D

Presentation Transcript

  1. Assimilation of Satellite Radiances into LM with 1D-Var and Nudging Reinhold, Christoph, Francesca, Blazej, Piotr, Iulia, Michael, Vadim DWD, ARPA-SIM, IMGW, NMA, RHM COSMO General Meeting, Cracow 15-19 September 2008 - plenary session -

  2. 1DVAR + Nudging = Nudgevar COSMO-Project:Assimilation of satellite radiances with 1D-Var and Nudging • Goals of Project: • Assimilate radiances (SEVIRI, ATOVS, AIRS/IASI) in COSMO-EU • Explore the use of nonlinear observation operators with Nudging • Explore the use of retrievals for regional models i.e. RETRIEVE temperature and humidity profiles and then nudge them as “pseudo”-observations

  3. Variational use of Satellite Radiances • Principle: • use model first guess (temperature and humidity profiles)‏ • simulate radiances from first guess (radiative transfer computation)‏ • adjust profiles until observed and simulated radiances match • - inversion by minimisation • - optimal merge of information • defined by observation and background errors • - keep vertical structure of model AMSU-A Temperature Weighting Functions Example: ATOVS of NOAA 15-18, METOP-A: 40 Channels (15 microwave, 19 infrared, 1 visible) Simulation based on 3-hour GME forecast Observation of NOAA 17, HIRS 8 (window channel)‏

  4. m/s Assimilation of satellite radiances with 1D-Var and Nudging Status: Slightly positive impact both for AMSU-A and SEVIRI... mean sea level pressure & max. 10-m wind gusts valid for 20 March 2007 , 0 UTC + 48 h, REF (no 1DVAR)‏ analysis AMSU-A: + 48 h, 1DVAR-THIN3 + 48 h, 1DVAR-THIN2 Reinhold Hess, 4

  5. Assimilation of satellite radiances with 1D-Var and Nudging ...but more tuning and long term trials are required for operational application • Activities during last COSMO-year: • Preparation of AMSU-Data from IMGW Centre, Processing from Database • Tuning of bias correction • Use of IFS forecast above model top instead of climate first guess • Tuning of observation error covariance matrix R • Tuning of background error covariance matrix B • Developments for IASI (cloud detection, bias correction, monitoring, tests)‏ • Still to be done: • ... Reinhold Hess, 5 Athens, 2007

  6. Assimilation of satellite radiances with 1D-Var and Nudging ...but more tuning and long term trials are required for operational application • Activities during last COSMO-year: • Preparation of AMSU-Data from IMGW Centre, Processing from Database • Tuning of bias correction • Use of IFS forecast above model top instead of climate first guess • Tuning of observation error covariance matrix R • Tuning of background error covariance matrix B • Developments for IASI (cloud detection, bias correction, monitoring, tests)‏ • Still to be done: • Thorough validation of Profiles • Further tuning of Nudging • Parallel Experiments, long term studies Reinhold Hess, 6

  7. Bias Correction for limited area model COSMO-EU • theoretical study (Gaussian error analysis): • two weeks of data is long enough for significant statistics sample size • predictors are highly correlated – chose representative synoptical and seasonal conditions Reinhold Hess, 7 Cost Funktion Variational Assimilation requires bias free observation increments H(x)-y bias from observation y, first guess x and radiative transfer H (RTTOV)‏ • bias correction in two steps: • remove scan line dependent bias • considered in H, however residual errors • remove air mass dependent bias • systematic errors related to • air mass temperature • air mass humidity • surface conditions • modeled with predictors • observed AMSU-4(5) and -9 • simulated AMSU-4 and 9 • model values, e.g. geop. thick, IWV, SST

  8. scanline biases AMSU/NOAA 18 (15 to 25 June 2007)‏ GME lat 30 to 60 deg, lon:-30 to 0 deg COSMO-EU: approx 1200-1500 fovs approx 1200 obs/fov approx 1000-1500 obs/fov Reinhold Hess, 8

  9. timeserie of bias corrected observations minus first guess AMSU-A channels 4-11, NOAA-16, ERA 40 stratosphere stable in the troposphere, however large variations for high sounding channels => use of channels AMSU-A 5-7 only Reinhold Hess, 9

  10. timeserie of bias corrected observations minus first guess AMSU-A channels 4-11, NOAA-16, IFS stratosphere stable in the troposphere, small variations for high sounding channels => use of channels AMSU-A 5-9 Reinhold Hess, 10

  11. Tuning of observation error covariance matrix R • Estimation of satellite observation-error statistics • in radiance space • with simulations based on radiosondes • intra-channel (vertical) correlations • horizontal correlations

  12. Tuning of background error covariance matrix B situation dependent vertical error structures derived from IFS blue: westerly winds red: stable high pressure scale dependent covariances with 500hPa correlations with 500hPa B defines the scales that are to be corrected flow dependent Idea: define B according to cloud classification SAF-NWC software for MSG1 and MSG2

  13. Developments for IASI: 8641 IR-channels (started in July 2007)‏ • cloud detection NWP-SAF McNally • bias correction (generalisation of bias correction predictors)‏ • upgrade to RTTOV-9 • monitoring (tartan/dns-plots)‏ • tests studies started‏ Time series (dna, tartan) of bias corrected o-b differences Analysis difference 500 hPa temperature [K] after 24 hours of assimilation

  14. Reinhold Hess, 14 COSMO Priority Project: Assimilation of Satellite Radiances with 1DVAR and Nudging Status of Developments September 2008 • technical implementation ready (ATOVS/SEVIRI/AIRS/IASI)‏ • basic monitoring of radiances (day by day basis)‏ • basic set up, case studies available • neutral to slightly positive results • stratospheric background with IFS forecasts • tuning of bias correction, R, B ww • To be done: • more nudging coefficients/thinning of observations required • long term evaluation • positive results • Use of 1D-Var developments available for other activities: • GPS tomography • Radar reflectivities

  15. Assimilation of satellite radiances with 1D-Var and Nudging Lessons learned: ->Climate first guess above model top has (negative) impact also for trophospheric channels ->Number of observations sufficient for bias correction, but representativity is issue ->Boundary values have a paramount impact on forecast quality, better use of observations in the centre of the models, quality of parameterisations ->Assimilation of clouds/humidity required ->Large scales hardly to be improved with radiances small scales and humidity to be improved Reinhold Hess, 15

  16. Thank You for attention Reinhold Hess, 16

  17. scanline biases AMSU/NOAA 18 (15 to 25 June 2007)‏ GME lat 30 to 60 deg, lon:-30 to 0 deg COSMO-EU: approx 1200-1500 fovs lapse rate? approx 1200 obs/fov approx 1000-1500 obs/fov Reinhold Hess, 17 Reading, 2007

  18. timeserie of bias corrected observations minus first guess AMSU-A channels 4-11, NOAA-18, ERA 40 stratosphere stable in the troposphere, however large variations for high sounding channels => use of channels AMSU-A 5-7 only Reinhold Hess, 18 Athens, 2007

  19. Reinhold Hess, 19 Cooperation with Vietnam: Application of 1D-Var and 3D-Var with HRM provide first guess values above model top (COSMO-EU: 30hpa)‏ • use of climatological values (ERA40) seems not sufficient • linear regression of top RTTOV levels from stratospheric channels • (other choice: use IFS forecasts as stratospheric first guess)‏ ECMWF profiles versus estimated profiles, top GME levels accuracy about 5K for lower levels, but ECMWF may have errors in stratosphere too levels: 0.10, 0.29, 0.69, 1.42, 2.611, 4.407, 6.95, 10.37, 14.81 hPa Athens, 2007

  20. Reinhold Hess, 20 Athens, 2007

  21. T-‘analysis increments’ from ATOVS, after 1 timestep (sat only), k = 20 no thinning of 298 ATOVS 30 ATOVS by old thinning (3)‏ 30 ATOVS, correl. scale 70% 40 ATOVS by thinning (3)‏ 82 ATOVS by thinning (2)‏ 82 ATOVS, correl. scale 70% Reinhold Hess, 21 Athens, 2007

  22. 1D-Var for LME – Cloud and Rain detection Microwave surface emissivity model: rain and cloud detection (Kelly & Bauer)‏ Validation with MSG imaging Validation with radar data Reinhold Hess, 22 Darmstadt, 2007

  23. courtesy: HIRLAM-DMI Reinhold Hess, 23 Reading, 2007

  24. Jan - 2003 - Feb courtesy: HIRLAM-DMI (Bjarne Amstrup)‏ Reinhold Hess, 24 Reading, 2007

  25. 1D-Var (compute each vertical profile individually): minimise cost functional temperature and humidity profile , first guess and error covariance matrix , observations (several channels) and error covariance matrix radiation transfer operator gives: The condition , analysed profile and analysis error covariance matrix The analysis is the mathematically optimal combination of first guess and observation given the respective errors Reinhold Hess, 25 Satellite Radiances – Developments at DWD for GME

  26. 1D-Var for LME – Assimilation of AMSU-A: Cloud and Rain detection Reinhold Hess, 26 Athens, 2007

More Related