slide1 n.
Skip this Video
Loading SlideShow in 5 Seconds..
Introduction to KENDA as COSMO Priority Project PowerPoint Presentation
Download Presentation
Introduction to KENDA as COSMO Priority Project

Loading in 2 Seconds...

play fullscreen
1 / 15

Introduction to KENDA as COSMO Priority Project - PowerPoint PPT Presentation

  • Uploaded on

Introduction to KENDA as COSMO Priority Project. Christoph Schraff Deutscher Wetterdienst, D-63067 Offenbach, Germany. KENDA : Km-scale ENsemble-based Data Assimilation. Motivation, implementation, status Current & future work.

I am the owner, or an agent authorized to act on behalf of the owner, of the copyrighted work described.
Download Presentation

PowerPoint Slideshow about 'Introduction to KENDA as COSMO Priority Project' - daphne-daniel

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.While downloading, if for some reason you are not able to download a presentation, the publisher may have deleted the file from their server.

- - - - - - - - - - - - - - - - - - - - - - - - - - E N D - - - - - - - - - - - - - - - - - - - - - - - - - -
Presentation Transcript

Introduction to KENDA as COSMO Priority Project

Christoph SchraffDeutscher Wetterdienst, D-63067 Offenbach, Germany

KENDA: Km-scale ENsemble-based Data Assimilation

  • Motivation, implementation, status
  • Current & future work

perturbations: LBC + IC + physics



Motivation : Why develop Ensemble-Based Data Assimilation ?

convection-permitting NWP:

after ‘few’ hours, a forecast of convection is a long-term forecast

  • deliver probabilistic (pdf) rather than deterministic forecast
  • need ensemble forecast and data assimilation system
  • (strategic aims in COSMO)

forecast component: COSMO-DE EPS

developed & operational at DWD

 ensemble-based data assimilation component missing & required

  • replace current nudging-based DA by state-of-the-art DA with flow-dependent B



x = 1 – 3 km

~ 2016 : x  2 km , LETKF

Motivation : Why develop Ensemble-Based Data Assimilation ?

 data assimilation: priority project within COSMO consortium

Km-scale ENsemble-based Data Assimilation (KENDA):








 Local Ensemble Transform Kalman Filter (LETKF, Hunt et al., 2007) ,

(because of its relatively low computational costs)



  • analysis for a deterministic forecast run : use Kalman Gain K of analysis mean

 deterministic run must use same set of observations as the ensemble system !

 deterministic run may have higher resolution (not optimal if deterministic f.g. deviates

strongly from ensemble mean f.g.)

xA = xB + K[yo – H(xB)]

LETKF (km-scale COSMO) :implementation

  • analysis step (LETKF) outside COSMO code

 ensemble of COSMO runs, collecting obs – f.g. 4D -LETKF

 separate analysis step code, LETKF included in 3DVAR package of DWD




Lateral BC /

other LETKF implementations

  • perturbed lateral BC : IFS EPS (MCH, ARPA-SIM)

(or at DWD) hybrid EnVar for ICON (GME)

variational formulation

(Buehner et al 2005)

high resolution



lower resolution



(40 members)

  • CNMCA (LucioTorrisiet al.) : LETKF for 10-km COSMO operational

implementation of LETKF features


  • main development of LETKF at DWD (Hendrik Reich , Andreas Rhodin),
  • main implemented features:
  • adaptive multiplicative covariance inflation (based on Desroziers statistics)
  • adaptive estimation of obs errors in obs space
  • adaptive estimation of obs errors in ensemble space (to account for limited Nens)
  • adaptive localisation to keep effective Nobs constant (to account for limited Nens)
  • multi-step analysis

implementation & LETKF tests

(so far using TEMP, aircraft, surface, wind profiler)

  • DWD:
      • stand-alone scripts for 2-day period: many LETKF tests, e.g. adaptive methods
      • LETKF in operational experimentation system NUMEX  slow (archive)
      • ‘BACY’ (basic cycling scripting environment for KENDA, Hendrik Reich):
        • fast (speed: DA with BACY ~ 1 – 2, i.e. ~ 5 – 10 times faster than with NUMEX)
        • largely portable (if obs / GME fields provided)
        • automatic plotting suite
        • model equivalent calculation (MEC) from forecasts for input to verification
        • potential: tool to ease collaboration with academia
  • scripting environments for LETKF DA cycle also at
    • MeteoSwiss: 1-hourly LETKF DA cycle for 1 month using conventional obs
    • ARPA-SIM: first tests, setting up OSSE (Chiara Marsigli)

KENDA : main short-term goal

Main aim: reach operationability in (mid/end) 2015

  • system complete (e.g. ana + perturb surface / soil) + robust + efficient
  • quality KENDA ≥ quality nudging-based opr. DA (incl. LHN) (deterministic)

(using similar obs set)

  • additional: provide IC perturbations for EPS
  • evaluation of EPS:
  • EPS: how to use KENDA IC perturbations for EPS (COSMO-DE-EPS)
      • (PP COTEKINO / Richard Keane, DWD)
      • replace or rather combine with current IC perturbations
      • HErZ LMU: structure & impact of KENDA IC perturbations (Florian Harnisch)
  • Diagnostics: FSO (forecast sensitivity of observations) (Matthias Sommer, LMU)

KENDA : short-term tasks

  • general testing, tuning, optimization of LETKF setup
    • specification of observation errors
    • use of adaptive methods (localisation, cov. inflation, R in ensemble space),
    • multi-step and multi-scale analysis with different obs / localisation scales
    • ensemble size (40 ?),
    • update frequency at ? RUC 1 hr  at  15 min ! (high-res. obs)

non-linearity vs. noise / lack of spread / 4D property ?

  • inclusion of additive covariance inflation,
      • probably using self-evolving perturbations (LucioTorrisi, CNMCA)
  • testing SPPT in DA cycle, possibly also perturbed physics parameters
  • inclusion of LHN (latent heat nudging) (as long as reflectivity not ready for use)
  • robustness: create new ensemble members, if few crash

Extended Use of Observations (1)

  • Aim: (implementation,) forecast improvements from using these observations
  • 3D radar radial velocity
      • Complete obs operator and efficient approximations suitable for DA developed,
      • thinning and superobbing strategies implemented, preliminary DA cycles
      • Yuefei Zeng, Uli Blahak (DWD)
      • (Status of Y. Zeng after June 2014 or other resources at DWD unclear)
  • 3D radar reflectivity (direct use)
      • Complete obs operator and efficient approximations suitable for DA developed,
      • thinning and superobbing strategies implemented, preliminary DA cycles
      • Virginia Poli,TizianaPaccagnella(ARPA-SIM);
      • Klaus Stephan (DWD), Theresa Bick (U. Bonn)

Extended Use of Observations (2)

  • GPS Slant Path Delay
      • Obs operators (incl. ray tracer) implemented in DWD global 3DVar;
      • Aim: implement complete and efficient obs operator in COSMO by end of 2014
      • Michael Bender ; ErdemAltunac(tomography) (DWD)
      • No resources available yet after 2014 for use in LETKF
      • (challenge to use horizontally + vertically non-local obs in LETKF)
  • Cloud Top Height (CTH) derived from Meteosat SEVIRI
      • Fully implemented, single-obs experiments, cycled DA with dense obs for low-stratus cases
      • Annika Schomburg(DWD, talk on Monday)
  • Direct use of SEVIRI IR window channels in view of assimilating cloud info
      • Obs operator (RTTOV) + data flow implemented, next monitoring + DA tests
      • Africa Perianez, DWD, until Feb. 2015, no resources yet thereafter
  • Exploratory: SEVIRI VIS/NIR window channels (Leonhard Scheck. LMU)

Extended Use of Observations (3) :


  • Mode-S (high-resolution) wind and temperature data (from aircraft)
      • and application to high-res airport model COSMO-MUC (with radar data)
      • Heiner Lange, TijanaJanjic-Pfander(HErZ LMU)
  • Screen-levelobservations (T-2m, q-2m, uv-10m)
      • (C. Schraff, DWD) (+ Master Thesis at MeteoSwiss on station selection)
  • Direct use of SEVIRI WV channels (for T, qv; for cloud info; linked to IR window)
      • Great interest by HErZ-LMU for a project, starting 2015



analysis members

LETKF for km-scale COSMO :method

  • implementation following Hunt et al., 2007
  • basic idea: perform analysis in the space of the ensemble perturbations
    • computationally efficient, but also restricts corrections to

subspace spanned by the ensemble

    • explicit localization (doing separate analysis at every grid point,

select only obs in vicinity and scale R-1)

    • analysis ensemble members

are locally linear combinations

of first guess ensemble members


KENDA : Analysis & Perturbation

of Lower Boundary Fields

  • Snow cover and depth, idea:

apply snow analysis independently to ensemble members (with perturbed obs ?)

  • Sea surface temperature (SST), idea: add perturbations to deterministic analysis
  • Soil moisture (soil temperature) perturbations only: as in EPS (COTEKINO)
  • Longer-term additional tasks
  • Soil moisture (soil temperature) analysis, by using screen-level obs; 2 ideas:
      • add 1 analysis level in LETKF for the soil, and
        • apply strong localization for calculating the transform matrix for this level
      • use the ensemble in current stand-alone variational SMA (perturbations ?)
  • Soil moisture analysis (+ perturbations) using satellite soil moisture data in LETKF
      • Eumetsat fellowship at CNMCA