1 / 24

Data Assimilation for Very Short-Range Forecasting in COSMO

Data Assimilation for Very Short-Range Forecasting in COSMO Christoph Schraff Deutscher Wetterdienst, Offenbach, Germany. operational : radar-derived precipitation rates by latent heat nudging in development : LETKF NWP for nowcasting : 2 examples. COSMO-DE : x = 2.8 km

adie
Télécharger la présentation

Data Assimilation for Very Short-Range Forecasting in COSMO

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. Data Assimilation for Very Short-Range Forecasting in COSMO Christoph SchraffDeutscher Wetterdienst, Offenbach, Germany • operational : radar-derived precipitation rates by latent heat nudging • in development : LETKF • NWP for nowcasting : 2 examples

  2. COSMO-DE: x = 2.8 km (deep convection explicit, shallow convection param.) ~ 2014 : x  2 km , LETKF COSMO consortium /convection permitting COSMO configurations Germany Greece Italy Poland Romania Russia Switzerland operational configurations : x = 2.2 – 2.8 km

  3. current COSMO DA:Observation Nudging Method: Dynamic Relaxation against observations ( : model state vector) G determines the characteristic time scale for the relaxation + assimilates high-frequency obs + continuous analyzed state  indirect obs  need retrievals  limited background error cross-covariances

  4. Required: relation: precipitation rate  model variables (observed) (info required by nudging) precipitation  condensation  release of latent heat • Approach: modify latent heating rates such that the model responds by producing the observed precipitation rates  Latent Heat Nudging (LHN) current COSMO DA: use of radar-derivedprecipitation by Latent Heat Nudging (LHN) • Assumption: vertically integrated latent heat release  precipitation rate

  5. LHN - temperature increment (in K/h) Scaling factor : Scaling factor : current COSMO DA: Latent Heat Nudging , implementation • Assumption: vertically integrated latent heat release  precipitation rate Vertical profiles: cloud liquid water content (in g/kg) latent heat release (in K/h)

  6. current COSMO DA: Latent Heat Nudging , general info LHN: modify temperature (latent heating) COSMO-DE: x = 2.8 km (deep convection explicit, shallow convection param.) + adjust specific humidity to maintain relative humidity • computationally efficient, applicable to complex microphysics • composite of precip rates every 5 min • adjustment applied locally in areas with precipitation, not in environment  strong, but short-lived positive impact radar composite as used since June 2011: 16 D, 2 NL, 2 B, 9 F, 3 CH , 2 CZ stations

  7. analysis + 1 h + 2 h + 3 h + 4 h + 5 h + 6 h current COSMO DA: Latent Heat Nudging , impact study x = 2.8 km , no convection parameterisation , LHN with humidity adjustment 1-hour sum of precipitation

  8. current COSMO-DE DA: LHN, scale-dependent verification 15 June – 15 July 2009 , 0-UTC COSMO-DE forecast runs FSS,30km (11 grid pts.) FSS,280km (101 g.p.) ETS,2.8km threshold 0.1 mm/h opr (LHN) no LHN 2.0 mm/h 5 10 15 20 5 10 15 20 5 10 15 20 forecast lead time [h]

  9. perturbations: LBC + IC + physics perturb. GME, IFS, GFS, GSM future (km-scale) COSMO DA:strategy 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 forecast component: COSMO-DE EPS pre-operational  products (precip beyond warning threshold) used by bench forecasters for lead times  3 hrs  ensemble-based data assimilation component required

  10. LETKF (COSMO) :method • COSMO priority project KENDA (Km-scale ENsemble-based Data Assimilation) • implementation following Hunt et al., 2007 • basic idea: do 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) • analysis ensemble members are locally linear combinations of first guess ensemble members

  11. deterministic run must use same set of observations as the ensemble system ! • Kalman gain / analysis increments not optimal, if deterministic background xB (strongly) deviates from ensemble mean background Analysis for a deterministic forecast run :use Kalman Gain K of analysis mean deterministic analysis recently implemented L : interpolation of analysis increments from grid of LETKF ensemble to (possibly finer) grid of deterministic run ensemble deterministic

  12. LETKF (km-scale COSMO) : scientific issues / refinement • ensemble size Nens = 32  40 • covariance inflation(adaptive multiplicative, additive) • localisation (multi-scale data assimilation, successive LETKF steps with different obs / localisation ? adaptive , dep. on obs density ? ) • update frequencyat ? 3 hr  RUC 1 hr  at  15 min ! non-linearity vs. noise / lack of spread / 4D property ? • perturbed lateral BC (ICON hybrid VAR-EnKF / EPS) noise control ? • non-linear aspects, convection initiation (outer loop , latent heat nudging ?) • technical aspects: efficiency, system robustness  2014 (quasi-)operational

  13. LETKF (km-scale COSMO) : some important observations at km scale • radar : direct 3-D radial velocity & 3-D reflectivity (start summer 2010) develop sufficiently accurate and efficient observation operators, soon available Particular issues for use in LETKF: obs error variances and correlations, superobbing, thinning, localisation

  14. LETKF (km-scale COSMO) : some important observations at km scale • ground-based GPS slant path delay (start Jan. 2012) • direct use in LETKF, or tomography • implement non-local obs operator in parallel model environment Particular issue: localisation for (vertic. + horiz.) non-local obs GPS stations (ZTD resp. IWV)

  15. LETKF (km-scale COSMO) : some important observations at km scale • cloud information based on satellite and conventional data (start March 2011) • derive incomplete analysis of cloud top + cloud base, using conventional obs (synop, radiosonde, ceilometer) and NWC-SAF cloud products from METEOSAT SEVIRI use obs increments of cloud or cloud top / base height or derived humidity

  16. LETKF (km-scale COSMO) : some important observations at km scale NWC-SAF SEVIRI cloud products: example cloud type CT cloud top height CTH fractional water clds high semitransparent very high clouds high clouds medium clouds low clouds very low clouds cloud-free water cloud-free land undefined COSMO: cloud water qc > 0 , or cloud ice qi > 5 .10-5 kg/kg clc = 100 % subgrid-scale clouds  clc = f(RH; shallow convection; qi ,qi,sgs) < 100 %

  17. LETKF (km-scale COSMO) : some important observations at km scale • cloud information based on satellite and conventional data (start March 2011) • derive incomplete analysis of cloud top + cloud base, using conventional obs (synop, radiosonde, ceilometer) and NWC-SAF cloud products from Meteosat SEVIRI use obs increments of cloud or cloud top / base height or derived humidity • use SEVIRI brightness temperature directly in LETKF in cloudy (+ cloud-free) conditions (in view of improving the horizontal extent of cloud / cloud top height) • compare approaches Particular issues: non-linear observation operators, non-Gaussian distribution of observation increments

  18. DWD nowcasting product with use of NWP : NowCastMIX , for storm prediction • displacement forecast: weighted mean using data from • KONRAD: radar-derived detection of storm cells + displacement vectors • CellMOS: displacement forecast based on radar / lightning data • RADVOR-OP: radar-derived forecast of precip + displacement • COSMO-DE: upper-air wind (?) • storm category using fuzzy logics • gust: COSMO-DE V-max (700 – 950 hPa) , displacement • rain: radar + fuzzy set based on KONRAD cell categ. , COSMO-DE PW , radar VIL • hail: radar VIL, KONRAD • lightning (yes / no)

  19. DWD nowcasting product with use of NWP : NowCastMIX example : forecast for next 90 min. thunderstorms with : gusts Bft 7 gusts Bft 8-10 gusts Bft 8-10, hail, heavy rain gusts Bft 8-10, hail, very heavy rain

  20. study on blendingprobabilistic nowcasting & NWP (EPS) Kober et al., 2011 probability of reflectivity > threshold (19 dBZ) nowcasting: by neighbourhood method (area grows at 1 km / minute,  240 km) + displacement (pyramidal optical flow technique, Keil and Craig, 2007) 2300 UTC: radar obs radar reflectivity at initial time of ‘forecast’ nowcast of probability valid for 14 July 2009, 2300 UTC

  21. Kober et al., 2011: blendingprobabilistic nowcasting & NWP (EPS) NWP probability: COSMO-DE-EPS N(Z>thr) / Nens (fraction method) (calibration with reliability diagram statistics)

  22. Kober et al., 2011: blendingprobabilistic nowcasting & NWP (EPS) seamless probabilistic blending additive combination in probability space

  23. Data Assimilation for very short-range forecasting in COSMO thank you for your attention

More Related