Eidgenössisches Departement des Innern EDI Bundesamt für Meteorologie und Klimatologie MeteoSchweiz Ensemble activites and plans at MeteoSwiss André Walser MeteoSwiss COSMO-GM, Rome, 5-9 September 2011
2 MeteoSwiss | COSMO-GM 2011 André Walser Overview Today: regional ensemble forecasts fully based on COSMO-LEPS Forecast ECMWF managed by ARPA-SIMC (time critical application) Reforecast ECMWF managed by MeteoSwiss Product generation & Swiss HPC center (CSCS) DMO Future: Development of COSMO-E
3 MeteoSwiss | COSMO-GM 2011 André Walser COSMO-LEPS clients and products Internal and ~20 external customers
4 MeteoSwiss | COSMO-GM 2011 André Walser Last year’s cloud cover problem… Low-level cloud cover (CLCL) ~50% of members produced low-level clouds
5 MeteoSwiss | COSMO-GM 2011 André Walser SYNOP verif T_2M CH JJA2010 ~2K Observations COSMO-LEPS ensemble mean COSMO-7 lead-time +(25-48)h 00 UTC
6 MeteoSwiss | COSMO-GM 2011 André Walser Last year‘s conclusion Cloud problem in COSMO-LEPS caused by Kain- Fritsch convection scheme together with a too moist boundary layer COSMO-EU soil merge should be introduced as soon as possible
7 MeteoSwiss | COSMO-GM 2011 André Walser Soil merge implementation Merge is operational since 11/04/ UTC forecast: fresh snow (FRESHSNW) snow density (RHO_SNOW) interception water (W_I) snow temperature (T_SNOW) snow water equivalent (W_SNOW) soil temperature (T_SO) soil moisture (W_SO) soil ice (W_ICE) Large impact on verification results for ensemble mean (following slides)
8 MeteoSwiss | COSMO-GM 2011 André Walser SYNOP verif T_2M CH JJA2011 Observations COSMO-LEPS ensemble mean COSMO-7 lead-time +(25-48)h 00 UTC
9 MeteoSwiss | COSMO-GM 2011 André Walser SYNOP verif T_2M Europe JJA2011 Bias STDE COSMO-LEPS ensemble mean outperforms COSMO-7 for all lead-times
10 MeteoSwiss | COSMO-GM 2011 André Walser SYNOP verif TD_2M Europe JJA2011 Bias STDE COSMO-LEPS ensemble mean outperforms COSMO-7 for all lead-times
11 MeteoSwiss | COSMO-GM 2011 André Walser SYNOP verif CLCT Europe JJA2011 Bias STDE COSMO-LEPS ensemble mean outperforms COSMO-7 for all lead-times
12 MeteoSwiss | COSMO-GM 2011 André Walser Soil merge: Impact on snow cover Snow height merged from COSMO-EU Snow height interpolated from IFS old suite New suite Large improvement due to higher resolution of snow analysis
13 MeteoSwiss | COSMO-GM 2011 André Walser COSMO-LEPS «wish list» Continue soil merge as long as COSMO-EU is in operation (up to 2014?) Keep COSMO model configuration close to COSMO-EU Implementation of 00 UTC forecast as soon as possible Use of „Super-Ensemble strategy“ for clustering still appropriate? Soil moisture perturbation
14 MeteoSwiss | COSMO-GM 2011 André Walser Status COSMO-LEPS reforecast suite reforecasts over a period of 20 years ( ) every 3 rd day with 90h lead time (~1 month in advance) same model version and configuration as COSMO-LEPS suite ERA-interim reanalysis as initial and boundary fields New: To account for COSMO-EU soil merge, a soil moisture bias correction is applied, based on a comparison of COSMO-EU and interpolated ERA- interim for 2008) Available at ECMWF mars archive Intention: statistically independent data set for a long period: retrieve, time=12:00:00, date= , stream=enfh, step=0/to/90, levtype=sfc, expver=1, type=cf, hdate= /to/ class=co, param= LT: 90h
15 MeteoSwiss | COSMO-GM 2011 André Walser Fieldextra news: Ensemble quantiles Fieldextra v10.4 allows to calculate user-defined quantiles, e.g.: &Process out_field = "TOT_PREC", product_category = 'eps_quantile', quantile = 75. / Simple algorithm (thanks to S. Theis for the input!): Q0 = ensemble min Q100 = ensemble max linear interpolation between two members
16 MeteoSwiss | COSMO-GM 2011 André Walser Fieldextra news: Ensemble quantiles Example: 5 member ensemble, equally weighted, calculate 25% quantile Sort members Calculate q values for each members (according to weights) Interpolate between members m3 m2m1 m5m4 * * * * * y(m x ) Q25 * q(m3)=10q(m2)=30q(m1)=50q(m5)=70q(m4)=90 Q40Q20
17 MeteoSwiss | COSMO-GM 2011 André Walser Development of COSMO-E
18 MeteoSwiss | COSMO-GM 2011 André Walser Expectations of our clients Focus on Alpine region / Alpine specific problems Consistency of products through different lead times and scales High availability of products Two broad classes of products: Very high resolution in space and time, typically 24 hours, rapid update cycle COSMO-1 Regional probabilistic forecasts up to 3-5 days COSMO-E
19 MeteoSwiss | COSMO-GM 2011 André Walser Boundary conditions for MeteoSwiss Human resources Limited: MeteoSwiss internal project COSMO-NExT (Novel Expert Tools) with no additional resources COSMO priority projects Collaboration with partners Computing resources National HPC strategy with funding for MeteoSwiss HP2C project for code adaptation to new architectures
20 MeteoSwiss | COSMO-GM 2011 André Walser Regional probabilistic forecasts: COSMO-E forecasts twice per day, ~3 km grid size, 60 levels VarEPS: Boundary conditions
21 MeteoSwiss | COSMO-GM 2011 André Walser COSMO-E setup initial conditions Implementation of Perturbations: LETKF with 40 members, PP KENDA ECMWF BC project COSMO-DE-EPS & PP CONSENS 21 ensemble members: final configuration depends on performance of COSMO code on our 2015 HPC system HP2C project / PP POMPA MeteoSwiss resources boundary conditionsmodel physics
22 MeteoSwiss | COSMO-GM 2011 André Walser Assimilation (LETKF, hourly update) Valid Time (UTC).. +5d Forecasts COSMO-E setup (2).. +5d forecast start ~75 min after analysis time (after COSMO-1) elapsed time ~45 min
23 MeteoSwiss | COSMO-GM 2011 André Walser Operational Time-table COSMO-LEPS COSMO-E Regular EPS runs LETKF Tests Preopera- tional Opera- tional setup physics & BC perturbations Verification & final tuning Setup LETKF
24 MeteoSwiss | COSMO-GM 2011 André Walser Thank you !