1 / 33

Y. Rochon, J.W. Kaminski, S. Heilliette, L. Garand, J. de Grandpr é, and R. Ménard

Observation system simulation experiments for the PREMIER mission Sub-task of the project ‘Quantification of Atmospheric Pollution and Climate Aspects’ (ESTEC c ontract No. 4000101294/10/NL/CBi). Y. Rochon, J.W. Kaminski, S. Heilliette, L. Garand, J. de Grandpr é, and R. Ménard

terica
Download Presentation

Y. Rochon, J.W. Kaminski, S. Heilliette, L. Garand, J. de Grandpr é, and R. Ménard

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. Observation system simulation experiments for the PREMIER missionSub-task of the project ‘Quantification of Atmospheric Pollution and Climate Aspects’ (ESTEC contract No. 4000101294/10/NL/CBi) Y. Rochon, J.W. Kaminski, S. Heilliette, L. Garand, J. de Grandpré, and R. Ménard Environment Canada 5h WMO Workshop on the Impact of Various Observing Systems on NWP Sedona, AZ, 22-25 May 2012

  2. Introduction • Objective: Acquire insight on the potential impact of PREMIER limb observations of temperature, water vapour and ozone in numerical weather (and ozone) prediction. • Applied methodology: The level of impact of PREMIER observations is estimated through Observation System Simulation Experiments (OSSEs). • Synthetic observation sets reflecting the characteristics of the projected PREMIER retrieval-type data and of existing observation systems are derived from a virtual truth, i.e. a nature run. • Target PREMIER-type data consists of retrieved profiles from the • InfraRed Limb Sounder (IRLS; T, water vapour, ozone, …) and • Millimetre-Wave Limb Sounder (MWLS; water vapour, ozone, …). • MLS-type data serves as additional benchmark for comparison. • Various assimilation and forecasting experiments with and without the PREMIER and MLS-type data are conducted and assessed.

  3. Nature Run • Chosen nature run is the T511NR provided by the European Centre for Medium-Range Forecasts (ECMWF) as contribution to the international Joint OSSE program (e.g. Reale et al., 2007; Masutani et al., 2008; http://www.emc.ncep.noaa.gov/research/JointOSSEs) • Free forecast run at triangular 511 spectral truncation (~40 km) and 91  levels with the top at 0.02 hPa (~80km) and ~0.4 km spacing in the UTLS • Observed 2005-06 sea surface temperatures and ice cover provided by the National Centers for Environment Prediction (NCEP) • Covers a 13-month period. The two time periods of this study are: 18 June 2005 to 31 August 2005 and 15 December 2005 to 28 February 2006 • Model configuration: same as that of the ECMWF IFS CY31r1 cycle • http://www.ecmwf.int/research/ifsdocs/CY31r1 • Ozone chemistry is done with the updated version of the Cariolle and Déqué (1986) parameterization

  4. Assimilation and forecasting system • NWP model: • Operational Global Environmental Multi-scale model (GEM) of Environment Canada • 800x600 (~33 km at 49) • 80 levels up to 0.1 hPa (0.3 to 0.6 km in the UTLS) • added linearized ozone chemistry (LINOZ; McLinden et al., 2000) • Initial conditions close to the NR (GEM 5-day forecasts from NR) • Assimilation approach: • 3D-VAR with FGAT (First Guess at Appropriate Time) • Incremental assimilation at T108 • Incremental control variables: , ’, T’, ln(q), O3, Ps • Global data assimilation every successive 6-hours • No background check applied for the OSSE assimilations • Surfaces conditions taken from CMC/EC (and not from the NR)

  5. Synthetic observations I • Sources of characteristics for the synthetic observation dataset (e.g. locations, types, numbers, spatial thinning): • Control dataset: • Real meteorological observations used at Environment Canada for Summer 2008 and Winter 2009 (transposed/relabelled to 2005/06) • Pre-thinned real datasets (post background checked data) for most meteorological observation sources • IR radiances: applied thinning in the simulation process. • SBUV/2 NOAA 17 and18 partial column ozone. • MLS temperature, water vapour and ozone profiles • PREMIER IRLS and MWLS observation characteristics

  6. Synthetic observations II • Observation simulations from the NR were done locally (at EC) using the various observation models already integrated in the assimilation system (including RTTOV8.7). • Result: assimilation-ready data in the format required for assimilation. A noise-free set was produced first. • IR brightness temperature simulations: (AIRS, IASI, GOES, …) • Observations simulated under cloud-affected and cloud-unaffected conditions (using NR cloud cover and ice/water liquid content) • Assimilation system not set to assimilate cloud affected radiances (equivalent brightness temperatures). • Thinning applied by removing cloud-affected brightness temp. and retaining only a cloud-unaffected value per 150 km x 150 km box.

  7. Meteorological control observations to be assimilated, excluding radiances (partly based on Laroche and Sarrazin, 2010)

  8. Control radiance observations IR

  9. PREMIER observationsTangent point orbit tracks for a sample 6 hour period (centered about 0 UTC) Across-tracks 1, 4, 7, 10 1429 5716 1748 584

  10. PREMIER observations (cont’d) • Vertical ranges and resolutions • - Minimum altitude: • - IRLS: up to 50 km with 1 km vert. resolution • MWLS: up to ~35 km with unequally spaced levels (>= 1.6 km) • Water dependent rejection conditions: • - affecting about 50% of IRLS and 30% of MWLS H2O of profiles in the lower tropospheric levels. • Averaging kernels: Not applied (except for SBUV-2 ozone) • - Current PREMIER averaging kernel matrices for T, H2O and O3 not used since they are nearly identity matrices in most (or all) of the vertical range • Observation error variances: • - Given random error variances plus added error variance offsets. • .

  11. Error standard deviations (NH extratropics) MLS IRLS (61/2 for some expts) MWLS

  12. Calibration: Observation random errors • Perturbations applied to the synthetic observations using Gaussian-distributed random errors. • Purpose of error level calibration: Provide greater confidence on the pertinence of the OSSE results. • Simple approach: • Desire statistical scores of 2/N (and its contributing terms) similar to those obtained from the assimilation of real observations, e.g. • Introduced error std. dev. scaling factors f (following Errico et al.) for different observation grouping (families) – derived here from ratios of the above two terms. • Limitations: • Adjustments (scaling factors) not dependent on vertical level • Some observation groupings contain sub-types (or channels) which may ideally require different adjustments • No application of spatial and inter-channel obs. error correlations.

  13. Control Control Real Real sk(xa)=2Jok/Nk Real obs. Jan. 2009 Control (synthetic) Winter/Summer Jan-Feb ‘06 Jul-Aug ‘05 f spectrally constant for this study not well calibrated

  14. Control vs real observation assimilation Comparison of 6h forecasts to radiosondes: (January) Real and synthetic Solid: mean difference Dashed: std. dev.

  15. Impact study results Global 6h forecast error levels (July) Solid: Mean error Dashed: Error std. dev. Black: Control Blue: Control+MLS Red: Control+IRLS U T q O3 Important caveat: max. iterations of 70 for most assim. expts (affects CRTL+IRLS assim. the most)

  16. Tropics 6h forecast error levels (July) Solid: Mean error Dashed: Error std. dev. Black: Control Blue: Control+MLS Red: Control+IRLS U T q O3

  17. U T South Pole 6h forecast error levels (July) Solid: Mean error Dashed: Error std. dev. Black: Control Blue: Control+MLS Red: Control+IRLS q O3

  18. Differences in RMS errors for temperature 6hr forecasts (July) [Control+MLS] – Control (negative values: RMSE reductions) RMSE for the Control [Control+IRLS] - Control

  19. Comparison of ratio of RMS errors (IRLS,Control)for temperature 6hr forecasts (July) TemperatureZonal wind component

  20. Time mean error differences of Control+IRLS minus Control for temperature 6hr forecasts (July)  =0.5 = 0.6

  21. Sample medium range forecast results for temperature (August) 6-day forecast time mean differences at  =0.2: [CTRL+IRLS] - CRTL Anomaly correlations (relative to the NR) CRTLCRTL+MLS CRTL+IRLS

  22. Time mean errors for water vapour 6hr forecasts (%; July) Control Control+MLS Control+IRLS Control+MWLS

  23. Ratio of RMS errors (IRLS, Control)for water vapour (lnq) 6hr forecasts (July)  =0.2 6-day forecasts

  24. Global mean and RMS errors, and anomaly correlation coefficients for water vapour (August) Source analyses: Control, Control+MLS, Control+IRLS Smaller due to cancelling of +/- errors

  25. Final remarks • Synthetic bias-free dataset reflecting realistic measurement network can be easily produced following the specification of the NR. • Zeroth order observation perturbation calibration feasible through a comparison of • Temperature (and winds): • RMS error reductions (zonal) usually within ~0.1-0.3 K and 0.2 m/s for IRLS and similar to half for MLS-type. • Increase in T predictability (avg over latitudes) of ~¼ (UTLS) to ½ (mid-strato) day. • More notable improvements near poles for T at about 1K • Improvement in T and U near 10 hPa at equator • Water vapour (and ozone): • IRLS (largest) and MWLS potential benefit over MLS-type in troposphere and UTLS (greater for ozone). • Error reductions extend downward to the mid-troposphere and below. • Present impact differences with MLS decrease with forecast length. • Notably persistent improvements over 10-day forecasts • Setup applicable to other studies.

  26. Acknowledgements • Nature run: ECMWF and the Joint OSSEs program (Michiko Masutani et al.). • Discussions on error calibration and setting of  =1 level of NR: Ronald Errico (GMAO/NASA and GESTC) • Discussions and information on PREMIER data: Lars Hoffmann, Bärbel Vogel, Joachim Urban, and Richard Siddans. • PREMIER Impact Study project management: Bärbel Vogel (Julich) and Joerg Langen (ESTEC) • MLS-Aura and SBUV/2 science and instrument teams for the availability of data. • Assimilation and forecasting system: Various EC colleagues

  27. Extras

  28. NR fields used Surface fields: Sea-ice cover, albedo, snow depth (to set snow cover field), skin temperature.

  29. Impact of PREMIER observations Some measures of performance • Monthly based RMS, std. dev. and time mean errors relative to the NR. • Ratios of RMS errors over individual months: (similarly to Lahoz et al., 2005) • A value of  <1 indicates a beneficial impact from X2 relative to X1. • Monthly mean differences (X2-NR) – (X1-NR) • Above accompanied by significance tests • Student t-test (mean diff.) anf F-test (applied to RMS errors) • Anomaly correlation coefficients relative to the NR.

  30. Comparison of ratio of RMS errorsfor ozone 6hr forecasts (July)  (IRLS, Control) (IRLS, MLS)  (MWLS, Control) (MWLS, MLS)

  31. RMS errors ozone 6hr forecasts (%; July) Control Control+MLS Control+IRLS Control+MWLS

  32. July time mean error ozone 6hr forecasts (%) Control Control+MLS Control+IRLS Control+MWLS

  33. Global mean errors, RMS errors, and anomaly correlation coefficients for ozone (August) Source analyses: Control, Control+MLS, Control+IRLS

More Related