470 likes | 574 Vues
The impact of moist singular vectors and ensemble size on predicted storm tracks for the winter storms Lothar and Martin. A. Walser 1) M. Arpagaus 1) M. Leutbecher 2) 1) MeteoSwiss, Zurich 2) ECMWF, Reading, GB. Moist vs. operational singular vectors Coutinho et al. (2004).
E N D
The impact of moist singular vectors and ensemble size on predicted storm tracks for the winter storms Lothar and Martin A. Walser1) M. Arpagaus1) M. Leutbecher2) 1)MeteoSwiss, Zurich 2)ECMWF, Reading, GB
Moist vs. operational singular vectorsCoutinho et al. (2004) • ‚opr‘ SVs (T42L31, OT 48 h): linearized physics package with • surface drag • simple vertical diffusion • ‚moist‘ SVs (T63L31, OT 24 h): linearized physics package includes additionally: • gravity wave drag • long-wave radiation • deep cumulus convection • large-scale condensation
Martin: Predicted storm tracks t+(42-66) < 980 hPa (1) ensemble members: 2 tracks ▬analysis • Configuration: • dry SVs/51 RMs • moist SVs/51 RMs < 970 hPa < 960 hPa
Martin: Predicted storm tracks t+(42-66) < 980 hPa (2) ensemble members: 12 tracks ▬analysis • Configuration: • dry SVs/51 RMs • moist SVs/51 RMs < 970 hPa < 960 hPa
Forecast storm Martin: max. wind gusts t+(30-54) (4) moist SVs, x~10 km • Configuration: • moist SVs/51 RMs • moist SVs/10 RMs
Evaluation of the COSMO-LEPS forecasts for the floods in Switzerland in August 2005 and further recent results… André Walser MeteoSwiss, Zurich
Case study: Flood event in Switzerland in August 2005 Photos:Tages-Anzeiger
Synoptic overview: 22 August 2005 Temperature 850 hPa and geopotential 500 hPa: 18º 10º 2º
Total precipitation over 3 days (20. – 22.8.) (06 - 06 UTC) C. Frei, MeteoSwiss About 400 stations, precipitation sum locally over 300 mm!
Probability precipitation > 100mm/72h C. Frei, MeteoSwiss
Probability precipitation > 250mm/72h C. Frei, MeteoSwiss
Summary • COSMO-LEPS has proved to provide useful forecast uncertainty estimates for extreme events. • In the case of the flooding of August, warning has been issued on 21st August. No early warning has been given on 19th, 20st August, although aLMo, LEPS gave correct signal. • The effect of previous false alarms should not be underestimated.
Use of Multi-Model Super-Ensemble Technique for complex orography weather forecast Massimo Milelli, Daniele Cane VII COSMO General Meeting Zuerich, September 20-23 2005
Multimodel Theory As suggested by the name, the Multimodel SuperEnsemble method requires several model outputs, which are weighted with an adequate set of weights calculated during the so-called training period. The simple Ensemble method with bias-corrected or biased data respectively, is given by (1) or (2) The conventional SuperEnsemble forecast (Krishnamurti et. al., 2000) constructed with bias-corrected data is given by (3)
Precipitation We use the following operational runs of the 0.0625° resolution version of LM (00 and 12 UTC runs) Local Area Model Italy (UGM, ARPA-SIM, ARPA Piemonte) (nud00, nud12) Lokal Modell (Deutscher Wetterdienst) (lkd00, lkd12) aLpine Model (MeteoSwiss) (alm00, alm12) Training: 180 days (dynamic) Forecast: from July 2004 to March 2005 Stations: 102 Method: mean and maximum values over warning areas
36-60 h Mean Maximum
Temperature We use the following operational runs of the 0.0625° resolution version of LM (00 and 12 UTC runs) Local Area Model Italy (UGM, ARPA-SIM, ARPA Piemonte) (nud00, nud12) Lokal Modell (Deutscher Wetterdienst) (lkd00, lkd12) aLpine Model (MeteoSwiss) (alm00, alm12) Training: 90 days (dynamic) Forecast: March 2005 Stations: 53 (h<700m), 34 (700m<h<1500m), 15 (h>1500m) Method: bilinear interpolation horizontally, linear vertically (using Z)
700 m < h < 1500 m h < 700 m h > 1500 m
h < 700 m 700 m < h < 1500 m h > 1500 m
Relative Humidity We use the following operational runs of the 0.0625° resolution version of LM (00 and 12 UTC runs) Local Area Model Italy (UGM, ARPA-SIM, ARPA Piemonte) (nud00, nud12) Lokal Modell (Deutscher Wetterdienst) (lkd00, lkd12) aLpine Model (MeteoSwiss) (alm00, alm12) Training: 90 days (dynamic) Forecast: March 2005 Stations: 53 (h<700m), 34 (700m<h<1500m), 15 (h>1500m) Method: bilinear interpolation horizontally, linear vertically (using Z)
h < 700 m 700 m < h < 1500 m h > 1500 m
Conclusions • Multimodel Ensemble and SuperEnsemble permit a strong improvement of all the considered variables with respect to direct model output. • In particular, SuperEnsemble is always superior to Ensemble, except for mean precipitation over warning areas and for ETS in general.
FIRST EXPERIMENTS FOR A SREPS SYSTEMChiara Marsigli ARPA-SIM - WG4
Basis: Operational COSMO-LEPS starting at 12 UTC of the 9th April 2005 (intense precipitation over Emilia Romagna region between 0 UTC of the 10 and 0 UTC of the 12 April 2005) • Set-up of the operational system: • model: LM version 3.9, no nudging, QI and prognostic precipitation • perturbations: b.c. and i.c. from 10 Representative Members selected out of 2 EPS; use of Tiedtke or Kain-Fritsch random • Experiment set-up: • Adding perturbations of the physical parameters of the model -> runtime perturbations • Forecast range: +72h
Interpretation of the new high-resolution model LMK Heike Hoffmann heike.hoffmann@dwd.de Volker Renner volker.renner@dwd.de Susanne Theis susanne.theis@dwd.de
Method We plan a two step approach 1. Using information of a single model forecast by applying the Neighbourhood Method (NM) 2. Using information resulting from LMK forecasts that are started every 3 h (LAF-Ensemble)
Neighbourhood Method Assumption: LMK-forecasts within a spatiotemporalneighbourhood are assumed to constitute a sample of the forecast at the central grid point
Products • Smoothed fields for deterministic forecasts • Expectation Values from spatiotemporal neighbourhood • simple averaging over quadratic grid boxes • Probabilistic Products • Exceedance Probabilities for certain threshold values for different parameters, especially for hazardous weather warnings
Verification results for precipitation Data LMK forecasts; 3.-17.01.2004; 13.-27.07.2004; 1 h values, 00 UTC and 12 UTC starting time; 7-18 h forecast time all SYNOPs available from German stations comparison with nearest land grid point Neighbourhood-Method-Parameters vers_01: 3 time levels (3 h); 10 s ( 28 km) vers_02: 3 time levels; 5 s vers_03: 3 time levels; 15 s Averaging square areas of different sizes (5x5,15x15)
Conclusions for deterministic precipitation forecasts • scores in winter are much better than in summer, but in summer there is more effect in postprocessing • expectation values of NM do not integrate into simple averaging, they show some advantages for intermediate thresholds • there is, however, no obvious overall improvement by using the NM instead of simple spatial averaging • therefore, simple averaging over 5x5 domain will be applied followed by a re-calibration of the distributions of the smoothed field towards the distribution of the original field
LMK Total Precipitation [mm/h] 13. Jan. 2004, 00 UTC, vv=17-18h [mm/h]
Exceedance Probability of 1mm/h, 13. Jan. 2004, 00 UTC, vv=17-18h [%] calculated with the NM with radius 10 grid steps, 3 time intervals (t-1, t, t+1)
Simple Kalman filter – a “smoking gun” of shortages of models? Andrzej Mazur Institute of Meteorology and Water Management
Introduction Model forecasts vs. observations Warsaw, Jan-Mar 2005`
“Raw” results vs. Kalman filtering - meteograms Application of simple Kalman filter for air temperature and wind speed (station Wroclaw)
“Raw” results vs. Kalman filtering - (post-processing) applications Application of simple Kalman filter for road temperature assessment during winter period
Conclusions Method seems to work quite good as far as “continuous” meteorological parameters, like temperature, wind speed or air pressure, are concerned. Other parameters, like precipitation, should be studied in a similar way. They might require different approach due to their different “nature”. In both cases, careful selection of predictors is strongly advised. Results also seem to depend on differences between observations and “raw” results (i.e., BEFORE filter is applied). The greater difference, the better result. Method - even in this simple approach - can “detect” not only any factor “aside” of the model, but also systematic errors in results.
WP4 Intrepretation and applications • WP4 intends to propose two projects • Short range ensemble with own perturbations (connexions with WG1, ensemble assimilation) • Interpretation, grid point statistics, input to forecast matrix, input to hydrological models