1 / 31

Trond Iversen

G rand L imited A rea M odel E nsemble P rediction S ystem. GLAMEPS - a multi-model EPS for the short range. Trond Iversen. A common project for operational EPS in the short-range in the HIRLAM and ALADIN SRNWP consortia, with strong links to ECMWF. Purpose of EPS: weather forcasting.

blake-french
Download Presentation

Trond Iversen

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. Grand Limited Area Model Ensemble Prediction System GLAMEPS - a multi-model EPS for the short range Trond Iversen A common project for operational EPS in the short-range in the HIRLAM and ALADIN SRNWP consortia, with strong links to ECMWF

  2. Purpose of EPS: weather forcasting • Ideal weather forecasts provide three elements: • The ”consensus” forecast: contains at any lead time all, and nothing more than, the predictable components; • Reliable forecast uncertainty of the ”consensus”; • Reliable probabilities of events relevant for individual users (with forecast resolution exactly reflecting predictability). • Predictability means: Reliable predictions with higher forecast resolution (sharper info) than climate data, are possible • Forecasts can be fully reliable at all forecast lead times (!) • Forecast resolution deteriorates with decreasing predictability, from ~1 (analysis) to ~0 (climate data)

  3. Is EPS needed for the short range? Development of ACC for Z(500hPa) for ECMWF’s Operatonal model: Development of ACC for Z(500hPa) for ECMWF’s Model used for re-analysis: Courtesy: P.Kållberg, A. Simmons; ECMWF

  4. Is EPS needed for the short range? Apparently, deterministic forecasts shorter than 1-2 days are close to perfect ! Courtesy: A. Simmons; ECMWF

  5. Is EPS needed for the short range? But perhaps less so for more direct weather parameters? Brier Skill Score for 96h EC EPS for selected events ff10m against anal 24hPrecip against anal 24hPrecip against obs ff10m against obs

  6. Predictability as a function of scale Except for occasional interactions between large-scale features and surface properties, small-scale events are rarely predictable beyond 1-2 days • Requires high spatial resolution with frequent updates • Requires very accurate and swiftly produced analyses • Large-scale flows, potentially embedding high-impact, weather can normally be predicted much longer Courtesy: A. Simmons; ECMWF

  7. Predictability as a function of scale Forecast lead time when RPSS for EC EPS of Z500 < 0.3 EPS determ ctrl All scales Planetary scales Sub-synoptic scales Synoptic scales

  8. Challenge extreme weather High-impact, extreme weather often involves a wide spectre of scales, e.g.: the larger scales provide conditions for potential occurrence But extremes often involve smaller scale, embedded structures (peak precip. and wind; fast temp. changes; etc.). Key issue: the time-window of predicting small-scale weather is short, and error saturation level is small the growth rate and saturation level of small scale errors is a potential limiting factor for predicting high-impact extremes Upscale loss of predictability is apparent: Errors grow on all scales, but saturate first for the small (except for interactions with fixed surface properties)

  9. ”Regression towards the mean”It is excessively optimistic to try to forecast rare events with single ”deterministic” predictions

  10. A ”time funnel” for reliable predictions Prediction Sharpness ~1 Initial- & Boundary-val problem mainly Initial-val problem boundary-val problem ~0 ? Lead time Climate range decadal range extended range medium range short range target 2-3 days 10-15 days month-year centuries 5-25(?) years

  11. Grand Limited Area Model Ensemble Prediction System The group ofGLAMEPS developers Project Leader: Inger-Lise Frogner(Trond Iversen until Dec 2011) Core personnel: K. Sattler (Hirlam), A. Deckmyn (Aladin) Calibration, presentation, verification: C. Santos, A. Deckmyn, G. Smeet, S. v.d.Veen, J. B. Bremnes, J. Kilpinen, M. Schmeits Operrational Aspects: X. Yang, U. Andraea Experimental developments: Hirlam SVs: J.Barkmeijer, S. v.d.Veen, ETKF: Å. Johansson, J. Bojarova Physics perturbations: H. Feddersen, A. Callado Thanks to ECMWF: Martin Leutbecher & Dominique Lucas

  12. Tellus 63A, No3, May 2011SR EPS special issue 4 papers of relevance for GLAMEPS

  13. Tests with pre-operational GLAMEPS_v0: 52 ensemble members; 13 per model . EuroTEPS (12 + 1) + HirEPS_K (12+1) + HirEPS_S (12+1) + AladEPS (13) = 52 In reality 51 unique members; AladEPS_00 is EuroTEPS_00 downscaled 13km grid resolution (Aladin 509x416,12.9km,L37); (Hirlam 486x378, 0.115deg,L40) Forecast range: 42h Multimodel approach: 2 versions of Hirlam (with different cloud and precipitation schemes 2 different LAMs (Aladin and Hirlam) 3 different analyses and control forcasts (EuroTEPS_00, HirEPS_K_00, HirEPS_S_00)

  14. First experimental results • Before upgrades of EC EPS in 2009, • initial and evolved SVs, T399 (~55km), stochastic physics • Test period, 00utc and 12utc,2008/0117 - 0308 • 4 experiments: • Control: A 52-member GLAMEPS • Alternatives: A 44-member GLAMEPS • Some examples with calibration (BMA, 3 days learning) • A range of 51-member, mono-model • Replacing EuroTEPS with operational EC EPS

  15. Talagrand diagrams, 2008/0117 - 0308 (00,12)MSLP +24h EPS_51 GLAMEPS_52 GLAMEPS_44 GLAMEPS_52 AladEPS_51 HirEPS_K_51 HirEPS_S_51

  16. Talagrand diagrams, 2008/0117 - 0308 (00,12)T2m +24h EPS_51 GLAMEPS_52 GLAMEPS_44 GLAMEPS_52 AladEPS_51 HirEPS_K_51 HirEPS_S_51

  17. Talagrand diagrams, 2008/0117 - 0308 (00,12)ff10m +24h EPS_51 GLAMEPS_52 GLAMEPS_52_BMA GLAMEPS_44 GLAMEPS_44_BMA GLAMEPS_52 AladEPS_51 HirEPS_K_51 HirEPS_S_51

  18. Brier Skill Score 2008/0117 - 0308 (00, 12 +24h) [ BSS = 1 – BSSReliab - BSSResol ] ff10m > 10m/s Pr6h > 5mm/6h EPS_51GLAMEPS_52 GLAMEPS_44 GLAMEPS_52_BMA GLAMEPS_44_BMA BSSResol BSSReliab GLAMEPS_52AladEPS_51 HirEPS_K_51 HirEPS_S_51 Multi-model vs.single model EPS of same size – no calibration

  19. Descrete Ranked probability skill score – DRPSS 2008/0117 - 0308 (00, 12) Using T399L62 EuroTEPS [ DRPSS = 1 - Reliability - Resolution ] T2m > -10, -5, 0, 5, 10, 15, 20, 25, 30C ff10m > 3, 5,10, 15, 20, 25, 30m/s Resolutionl EPS_51, GLAMEPS_44 GLAMEPS_52, GLAMEPS_52-BMA Reliability Multi-model vs.single model EPS of same size – no calibration GLAMEPS_52AladEPS_51 HirEPS_K_51 HirEPS_S_51

  20. DRPSS 12-42h, 6h Precip Multi-model vs.single model EPS of same size – no calibration Pr6h > 0.1, 1, 2, 5, 10, 15, 20, 25, mm/6h

  21. BSS, Rare events – BMA calibration for ff10m (3-days calibration) Ff10m > 20m/s Prec > 10mm/6h Calibration is not straightforward for rare events and heterogeneous climatology. EPS_51GLAMEPS_52 GLAMEPS_44

  22. Pre-operational prod EuroTEPS replaced by operational EC EPS EuroTEPS Operational 51-member EC EPS, T639L62 & EDA Verification of 52-member GLAMEPS compared with operational EC EPS DRPSS 12-42h T2m ff10m Aug-Sept-Oct 2010

  23. Pre-operational prod EuroTEPS replaced by operational EC EPS Aug-Sept-Oct 2010 DRPSS 12-42h, 6h Precip

  24. Summary on _v0 • Clear improvement over operational EC EPS • Also after upgrades of EC EPS with higher resolution and inclusion of Ensemble Data-Assimilation (EDA) • Multi-model better than single model EPS • Exceptions exist: systematic un-even model quality • Model-specific bias-correction may help • Replacing EuroTEPS with EC-EPS • Degrades, but only very slightly • After upgrades of EC EPS, the extra cost of running EuroTEPS is better used to improve other aspects of GLAMEPS

  25. The design of the _v1 • Replace EuroTEPS with 51-member EC EPS • Utilize all 51 ensemble members • HirEPS and AladEPS with ~11-11.5 km on 30% larger domains • for longer lead-times (54h) starting from 06 and 18 utc • Aladin and Hirlam upgraded to latest versions • Multiple surface DA in Aladin and Hirlam (SURFEX + CANARI). • SMS-scripts for operational prod at ECMWF (TCF Opt 2): • Routines for production monitoring • Operational emergencies RT-actions • Products and presentations • Raw data for download, Grib 2 • Operational verification Hirlam-domain Aladin-domain

  26. GLAMEPS_v1 planned production schematic

  27. https://glameps.org Pre-Operational GLAMEPS For Example Forecasts, valid 15/05/2012 12utc Z500 ensemble mean + RMS spread Prob [Prec > 0.5mm/3h] +30h +30h +54h +54h

  28. Pre-Operational GLAMEPS: GLAMEPS-o-grams for Reading, https://glameps.org from 13/05/2012 18utc from 14/05/2012 06utc EC EPS-gram from 14/05/2012 00utc

  29. Pre-Operational GLAMEPS: selected verification: crpss relative to EC EPS ff10m +36h, +65 deg N T2m +36h, +65 deg N prec24h +24-48hh, +65 deg N prec24h +24-48hh, Europe ff10m +36h, Europe T2m +36h, Europe

  30. Furtherdevelopment LAM-specific SVs • CAPE , CAPE suppressing, or other LAM SVs in HIRLAM • Investigate role of diabatic processes, resolution, and optimization time • Blending with larger scale perturbations ETKF, EDA, Hybrid etc. • Further studies on inflation factors due to small ensemble size • blending with larger scale perturbations • Comparison between EDA-Hybrid and ETKF-Hybrid • Ensembles are used to etimate flow-dep. model error covariance (“B-matrix”) Introduce higher-resolution ensemble members in GLAMEPS

  31. Thank You&Good Luck!

More Related