13 th SRNWP / 28 th EWGLAM Meeting Zürich, 9 – 12 Oct 2006 1 christoph.schraff@dwd.de • current status • long-term strategy • mid-term strategy • some ongoing or planned activities Overview and Strategy on Data Assimilation for LM [email protected]Deutscher Wetterdienst, D-63067 Offenbach, Germany Jürgen Steppeler CH , D , GR , I , PL , RO ( cosmo-model.cscs.ch )
13
Embed
current status long-term strategy mid-term strategy some ongoing or planned activities
CH , D , GR , I , PL , RO ( cosmo-model.cscs.ch ). Overview and Strategy on Data Assimilation for LM [email protected] Deutscher Wetterdienst, D-63067 Offenbach, Germany Jürgen Steppeler. current status long-term strategy mid-term strategy some ongoing or planned activities. - PowerPoint PPT Presentation
Welcome message from author
This document is posted to help you gain knowledge. Please leave a comment to let me know what you think about it! Share it to your friends and learn new things together.
• PDFs: deliver not only deterministic forecasts, but a representation of the PDF (ensemble members with probabilities), particularly for the convective scale
• use of indirect observations at high frequency even more important
Generalized global + regional FC + DA: ICON (DWD + MPI)
• global non-hydrostatic model with regional grid refinement for - global and regional modelling
- NWP and climate
• will replace GME and LM-E in 2010& provide lateral boundaries for convective-scale LM-K
• 3DVAR with Ensemble Transform Kalman Filter
Long-term strategy
emphasis on ensemble techniques (FC + DA)
due to special conditions in convective scale (non-Gaussian pdf, balance flow-dependent and not well known, high non-linearity), DA split up into:
– generalised DA for global + regional scale modelling ( variational DA)
– separate DA for convective scale
Data Assimilation for LM: Long-term Vision & Strategy
SIR method can handle the major challenges on the convective scale:
• Non Gaussian PDF• Highly nonlinear processes • Model errors• Balance (unknown and flow-dependent)• Direct and indirect observations with highly nonlinear observation operators and norms
• COSMO: gets lateral b.c. from LM-SREPS, provides initial conditions for LM-K EPS
Data Assimilation for LM: Long-term Strategy for Convective Scale
Potential problems: Ensemble size, filter can potential drift away from reality, but it cannot be brought back to right track without fresh blood,dense observations may not be used optimally
However:
• for LM-K: Strong forcing from lower and lateral boundaries expected to avoiddrift into unrealistic states
• if method does not work well the pure way: Fallback positions:– combine with nudging: (some) members be (weakly) influenced by nudging– approaches for localising the filter
• start development of SIR (for the longer-term, with option to include nudging)
Data Assimilation for LM: Mid-term Strategy
• Nudging at moment: – robust and efficient– requires retrievals for use of indirect observations– no severe drawbacks (for short term, convective scale)
if we can make retrievals available
→ further develop nudging, in particular retrieval techniques
• derive 3-dim. wind field from 3 consecutive scans of 3-d reflectivity and radial velocity at 10’-intervals, by means of a simple adjoint (SA) method (ARPS, Gao et. al. 2001)
• Cost function with 2 observation terms :
1. for radial velocity: in a standard way
2. for a tracer (reflectivity): reflectivity from 1st scan advected with the retrieved velocity and compared to reflectivity observations from 2nd and 3rd scan
horizontal wind retrievalDoppler radial wind at 2000 m , 13:04 UTC
[km
]
Legionowo
(Warsaw)
Radar
26-07-2003
Data Assimilation for LM: Radar Data: Simple Adjoint 3-D Wind Retrieval (PL)
• recently: noise problems for real data from Polish radars much reduced, method works now for single doppler radar