Deutscher Wetterdienst COSMODEEPS Susanne Theis Christoph Gebhardt Michael
Deutscher Wetterdienst COSMO-DE-EPS Susanne Theis, Christoph Gebhardt, Michael Buchhold, Zied Ben Bouallègue, Roland Ohl, Marcus Paulat, Carlos Peralta with support by: Helmut Frank, Thomas Hanisch, Ulrich Schättler, etc
NWP Model COSMO-DE è grid size 2. 8 km è without parametrization of deep convection (convection-permitting) è lead time 0 -21 hours è operational since April 2007 COSMO-DE COSMO-EU GME COSMO GM – September 2010
Plans for a COSMO-DE Ensemble How many ensemble members? è preoperational: 20 members è operational: 40 members When? è preoperational: 2010 è operational: COSMO GM – September 2010 2012
COSMO-DE-EPS production steps Ensemble products: „variations“ within forecast system ensemble members COSMO GM – September 2010 - mean - spread - probabilities - quantiles -. . .
COSMO-DE-EPS production steps Ensemble products: „variations“ within forecast system ensemble members + verification + postprocessing COSMO GM – September 2010 - mean - spread - probabilities - quantiles -. . .
COSMO-DE-EPS production steps „variations“ within forecast system Ensemble products: 1 ensemble members next slides: step 1, production of members COSMO GM – September 2010 - mean - spread - probabilities - quantiles -. . .
Generation of Ensemble Members Variations in Forecast System for the Representation of Forecast Uncertainty Initial Conditions COSMO GM – September 2010 Boundaries Model Physics
Generation of Ensemble Members Variations in Forecast System for the Representation of Forecast Uncertainty Initial Conditions Boundaries “multi-model” driven by different global models COSMO GM – September 2010 Model Physics
Generation of Ensemble Members Variations in Forecast System for the Representation of Forecast Uncertainty Initial Conditions Boundaries “multi-model” modification of COSMO-DE initial conditions by different global models driven by different global models COSMO GM – September 2010 Model Physics
Generation of Ensemble Members Variations in Forecast System for the Representation of Forecast Uncertainty Initial Conditions Boundaries Model Physics “multi-model” “multi-configuration” modification of COSMO-DE initial conditions by different global models driven by different global models different configurations of the COSMO-DE model COSMO GM – September 2010
Generation of Ensemble Members The Ensemble Chain global COSMO 7 km COSMO GM – September 2010 COSMO-DE 2. 8 km COSMO-DE mm/24 h
Generation of Ensemble Members The Ensemble Chain global COSMO 7 km COSMO GM – September 2010 plus the variations of • initial conditions • model physics COSMO-DE 2. 8 km COSMO-DE mm/24 h
Generation of Ensemble Members è Which computers are used? è at ECMWF: „ 7 km Ensemble“ è at DWD: COSMO-DE-EPS global COSMO 7 km ECMWF COSMO GM – September 2010 COSMO-DE 2. 8 km DWD
Generation of Ensemble Members è Which computers are used? COSMO 7 km è at ECMWF: „ 7 km Ensemble“ è at DWD: COSMO-DE-EPS GME IFS transfer of data COSMO GM – September 2010 GFS …etc…
Generation of Ensemble Members è Which computers are used? COSMO 7 km è at ECMWF: „ 7 km Ensemble“ è at DWD: COSMO-DE-EPS è Status: in testing phase GME IFS (so far: COSMO-SREPS) transfer of data COSMO GM – September 2010 GFS …etc…
store time minus initial time [hours] Generation of Ensemble Members SREPS data in DWD database 9 8 7 6 5 4 July / August 2010 COSMO GM – September 2010 transfer of data
Generation of Ensemble Members è variation of initial conditions COSMO GM – September 2010
Generation of Ensemble Members è variation of initial conditions global forecasts COSMO GM – September 2010
Generation of Ensemble Members è variation of initial conditions ic global forecasts COSMO 7 km COSMO-DE 2. 8 km COSMO GM – September 2010
Generation of Ensemble Members COSMO-DE assimilation è variation of initial conditions IC ic global forecasts COSMO 7 km COSMO-DE 2. 8 km COSMO GM – September 2010
Generation of Ensemble Members COSMO-DE assimilation è variation of initial conditions IC modify initial conditions of COSMO-DE by using differences between the COSMO 7 km initial conditions IC´ = F (IC, ic – icref) ic global forecasts COSMO 7 km COSMO-DE 2. 8 km COSMO GM – September 2010
Spread vs Lead Time IC pert Oct 7 – Nov 24 2009 15 days selected (15 ensemble members) Measure of ensemble spread: interquartile range of precipitation COSMO GM – September 2010
Spread vs Lead Time (Case Study) Case Study 18 June 2009 (15 ensemble members) IC pert convective event Measure of ensemble spread: interquartile range of precipitation COSMO GM – September 2010
Generation of Ensemble Members è variation of „model physics“ COSMO GM – September 2010 different configurations of COSMO-DE 2. 8 km: 1 entr_sc 2 rlam_heat 3 rlam_heat 4 q_crit 5 tur_len
Generation of Ensemble Members è variation of „model physics“ Selection of Configurations subjective, based on experts, verification Selection Criteria: different configurations of COSMO-DE 2. 8 km: 1 entr_sc 2 rlam_heat 3 rlam_heat 4 q_crit 5 tur_len 1. large effect on forecasts 2. no „inferior“ configuration COSMO GM – September 2010
Generation of Ensemble Members è no perturbations of soil moisture (so far) some sensitivity experiments COSMO GM – September 2010
Soil moisture sensitivity experiments è In IC+PH+BC exp no perturbations at the surface è Use COSMO-EU SO analysis fields: W_SO and/or T_SO è Complete or partially replace SO from COSMO-DE with COSMO-EU fields: WSO_DE = WSO_DE ± C (WSO_EU – WSO_DE), C = 0. 5. . 1 è Start at 06 UTC, 24 hours of forecast è Currently only single runs (no ensemble mode). è Verification for July 2009: Quality of forecast does not decrease COSMO GM – September 2010
COSMO-DE SO exp 1 cm 6 cm Overall structure similar. Localized differences (rel diffs ~10 %). COSMO-DE does a good job by itself. COSMO GM – September 2010
COSMO-DE SO Exp Obs. Fcast does not improve (not the point). Want to increase uncertainty range (without being blatantly wrong). COSMO GM – September 2010
Generation of Ensemble Members Variations within Forecast System: Model Physics 20 Ensemble Members (here still with COSMO-SREPS) (this design will be modified) Initial Conditions & Boundaries 1 IFS GME GFS UM COSMO GM – September 2010 2 3 4 5 Lhn_coeff=0. 5
Case Study: 18 June 2009 Lead times: 0 -6 hours 6 -12 hours 12 -18 hours 18 -24 hours mm/6 h COSMO GM – September 2010
Impact of different initial & boundary conditions Derived by global model: IFS (ECMWF) GME (DWD) GFS (NCEP) mm/24 h COSMO GM – September 2010
Impact of different initial & boundary conditions Derived by global model: IFS (ECMWF) GME (DWD) GFS (NCEP) mm/24 h COSMO GM – September 2010
Impact of different model configurations Derived by changing: entr_sc rlam_heat tur_len mm/24 h COSMO GM – September 2010
Impact of different model configurations Derived by changing: entr_sc rlam_heat tur_len mm/24 h COSMO GM – September 2010
Generation of Ensemble Members è future changes - extension to 40 members - switch to ICON as driving ensemble (model ICON currently under development) - apply an Ensemble Kalman Filter for initial condition perturbations (En. KF currently under development for data assimilation) COSMO GM – September 2010
COSMO-DE-EPS production steps „variations“ within forecast system 2 ensemble members next slides: step 2, generating „products“ COSMO GM – September 2010 Ensemble products: - mean - spread - probabilities - quantiles -. . .
Generation of „Ensemble Products“ è variables (list will be extended): è 1 h-precipitation 2 è wind gusts è 2 m-temperature GRIB 1 ensemble products: è ensemble „products“: è probabilities è quantiles è ensemble mean è min, max è spread COSMO GM – September 2010 - mean - spread - probabilities - quantiles -. . . GRIB 2
Generation of „Ensemble Products“ è further improvement: è adding a spatial neighbourhood è adding simulations started a few hours earlier COSMO GM – September 2010
Generation of „Ensemble Products“ è further improvement: è adding a spatial neighbourhood è adding simulations started a few hours earlier è additional product: probabilities event somewhere in 2. 8 km Box event somewhere in 28 km Box with upscaling % COSMO GM – September 2010
Product Generation: Example Probability of Event „Precipitation > 10 mm/24 h“ % COSMO GM – September 2010
Product Generation: Example Probability of Event „Precipitation > 10 mm/24 h“ as usual Derived from: Ensemble Members + Spatial Neighbourhood Defined for: area of 10 x 10 Grid boxes % COSMO GM – September 2010
COSMO-DE-EPS production steps Ensemble products: „variations“ within forecast system ensemble members - mean - spread - probabilities - quantiles -. . . 3 next slides: step 3, visualization in Nin. Jo COSMO GM – September 2010
Visualization in Nin. Jo è New Development: „Ensemble Layer“ è for Nin. Jo Version 1. 3. 6 è released in 2010 COSMO GM – September 2010
Visualization in Nin. Jo Screenshots COSMO GM – September 2010
Visualization in Nin. Jo Selection: Ensemble-Layer COSMO GM – September 2010
Visualization in Nin. Jo Selection: Ensemble Prediction System, Forecast Time, Variable COSMO GM – September 2010
Visualization in Nin. Jo Selection: Ensemble Prediction System, Forecast Time, Variable COSMO GM – September 2010
Visualization in Nin. Jo Selection: “Ensemble Product” COSMO GM – September 2010
Visualization in Nin. Jo Example: Probability of precipitation > 1 mm COSMO GM – September 2010
Visualization in Nin. Jo Selection: Meteogram COSMO GM – September 2010
Visualization in Nin. Jo Example: Meteogram (Quantiles) Zeit COSMO GM – September 2010
COSMO-DE-EPS production steps Ensemble products: „variations“ within forecast system ensemble members COSMO GM – September 2010 - mean - spread - probabilities - quantiles -. . .
COSMO-DE-EPS production steps Ensemble products: „variations“ within forecast system ensemble members + verification + postprocessing COSMO GM – September 2010 - mean - spread - probabilities - quantiles -. . .
Verification of COSMO-DE-EPS è quality of single members - deterministic verification è quality of ensembles - probabilistic verification of probabilities - ensemble spread COSMO GM – September 2010
Verification Results è Very first aim: Does the ensemble meet some basic requirements? è Results: - ensemble spread is present - members are of similar quality - quality of ensemble is superior to quality of individual forecasts GEBHARDT, C. , S. E. THEIS, M. PAULAT, Z. BEN BOUALLÈGUE, 2010: Uncertainties in COSMO-DE precipitation forecasts introduced by model perturbations and variation of lateral boundaries. Submitted to Atmospheric Research. COSMO GM – September 2010
Talagrand Diagram Oct 7 – Nov 24 2009 15 days selected (15 ensemble members) COSMO GM – September 2010
Individual Member Verification ETS COSMO GM – September 2010 FBI Oct 7 – Nov 24 2009 15 days selected (15 ensemble members)
Probabilistic Verification of Ensemble Brier Skill Score 0. 4 0. 3 0. 2 0. 1 0. 0 reference: deterministic COSMO-DE threshold in mm/h COSMO GM – September 2010 Oct 7 – Nov 24 2009 15 days selected (15 ensemble members)
Probabilistic Verification of Ensemble è Ph. D Andreas Röpnack (DWD, Univ of Bonn) è developed new verification measure (Bayesian) è applies it to COSMO-DE-EPS and COSMO-SREPS COSMO GM – September 2010
Postprocessing / Calibration COSMO GM – September 2010
Postprocessing / Calibration Ensemble products: „variations“ within forecast system ensemble members COSMO GM – September 2010 - mean - spread - probabilities - quantiles -. . .
Postprocessing / Calibration Ensemble products: „variations“ within forecast system ensemble members COSMO GM – September 2010 - mean - spread - probabilities - quantiles -. . .
Motivation for Postprocessing / Calibration è Aim: improve the quality historical data è learn from past forecast errors Forecast Obs è derive statistical connections è apply them to real-time ensemble forecasts real-time forecasts COSMO GM – September 2010
Methods for Postprocessing / Calibration è First Approach: statistical postprocessing logistic regression ensemble „products“: - mean - spread - probabilities - quantiles -. . . COSMO GM – September 2010
Methods for Postprocessing / Calibration è First Approach: logistic regression Forecast of Probabilities by statistical connection between Predictor Observation Yes / No Probabilities 1. 0 0. 5 0. 0 0 1 2 3 5 Predictor 8 12 e. g. Ensemble mean [mm] COSMO GM – September 2010
Methods for Postprocessing / Calibration è First Approach: logistic regression Forecast of Probabilities by statistical connection between Predictor Observation Yes / No Plan: preoperational in 2011 for precipitation COSMO GM – September 2010 Probabilities 1. 0 0. 5 0. 0 0 1 2 3 5 Predictor 8 12 e. g. Ensemble mean [mm]
Research for Postprocessing / Calibration è In addition: Research at Universities, funded by DWD - University of Bonn: Petra Friederichs, Sabrina Bentzien Methods: Quantile Regression, Extreme Value Statistics - University Heidelberg: Tilmann Gneiting, Michael Scheuerer Methods: Bayesian Model Averaging, Geostatistics COSMO GM – September 2010
COSMO-DE-EPS production steps Ensemble products: „variations“ within forecast system ensemble members + verification + postprocessing COSMO GM – September 2010 - mean - spread - probabilities - quantiles -. . .
Plans COSMO-DE-EPS è 2010: start of preoperational phase (20 members) è 2010 -2012: further extensions è statistical postprocessing è 40 members è 2012: start of operational phase convection-permitting ensemble operational COSMO GM – September 2010
- Slides: 70