1 / 52

Sensitivity of AROME Model to Modified Semi-Lagrangian Scheme

Evaluation of the AROME model's performance at convective scale with a modified semi-Lagrangian scheme. Includes tests of precipitation bias and convection behavior.

mahaney
Télécharger la présentation

Sensitivity of AROME Model to Modified Semi-Lagrangian Scheme

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. WWOSC 2014 • 16-21 August, Montréal, Canada • Les règles générales • Didier Ricard1, Sylvie Malardel2, Yann Seity1 • Julien Léger1, Mirela Pietrisi • 1. CNRM-GAME, METEO-France, Toulouse • 2. ECMWF, Reading • Sensitivity of short-range forecasting with the AROME model to a modified semi-Lagrangian scheme and high resolution.

  2. 1 – Introduction • AROME (Seity et al., 2011): operational fine-scale NWP model used at METEO-France since 2008 • In 2008: 2.5-km horizontal resolution, 41 vertical levels Domain 1500 km * 1300 km (600*512 points) • Current version: 2.5-km horizontal resolution, 60 vertical levels Domain 1875 km * 1800 km (750*720 pts) • In 2015: 1.3-km horizontal resolution, 90 vertical levels Domain 1996 km * 1872 km (1536*1440 pts) • Next version: 1.3 km 90

  3. Characteristics of the AROME model 1 – Introduction • Dynamics package: • Nonhydrostatic model based on a fully compressible system • Spectral model, A grid • Semi-Lagrangian scheme • Tri-linear interpolation for computation of trajectories (origin point) • quasi-cubic interpolations for calculating advected variables at origin point • Time scheme • 2 Time Levels semi-implicit scheme with SETTLS option (operational version) • ICI (iterative centred implicit) scheme (Predictor-corrector scheme) • 4th order spectral diffusion and gridpoint SLHD on hydrometeors • Physics package: • one moment mixed-phase microphysical scheme: 5 hydrometeor classes • 1D Turbulence scheme: pronostic TKE equation with a diagnostic mixing length (Bougeault Lacarrere, 1989) • Surface scheme: SURFEX (ISBA parametrisation, TEB scheme for urban tiles, ECUME for sea tiles) • Radiation scheme: ECMWF parameterization • EDMF Shallow convection scheme

  4. 1 – Introduction  Evaluation of the AROME model at convective scale for preparing the next operational version • Test of a modified SL scheme at 2.5-km horizontal grid spacing during several periods (in particular between 15 July - 15 September 2013) • Comparison between AROME forecasts at 1.3-km and 2.5-km horizontal resolutions during June-November 2012 for days with thunderstorms

  5. Motivation 2 – Test of a modified Semi-Lagrangian scheme • Bias for precipitation: • too much precipitation • sometimes too strong outflows under convective cells (with a strong diffusion) • Convection: • small-scale processes dominated by divergent modes • strong interaction between physics and dynamics • excessive behaviour: lack of conservation of SL scheme is suspected • Solution: more conservative SL schemes (CISL, finite volume …) • complex to implement • expensive for operational use • Simpler alternative approach (proposed by S. Malardel): • taking into account expansion/contraction of atmospheric parcels associated to each gridpoint • small modifications of the SL interpolation weights as a function of deformation •  Evaluation on a 2-month period (15 July 2013 - 15 September 2013) including deep convection with important effects of divergence

  6. COMAD scheme (Malardel and Ricard, in review, QJ) 2 – Test of a modified Semi-Lagrangian scheme • * Computation of the trajectories: no modification • t+1 • t • O • Departure or origin point

  7. COMAD scheme (Malardel and Ricard, in review, QJ) 2 – Test of a modified Semi-Lagrangian scheme • * Computation of the trajectories: no modification • * Computation of the value of variables at the origin point  modification of the SL interpolation weights • For example, with linear interpolations (2D and regular grid): • Original SL scheme: • 2 linear zonal interpolations • VB = wx1 VB1 + wx2 VB2 with wx2 =  /dx, wx1 = 1 -  /dx • VC = wx1 VC1 + wx2 VC2 = 1 - wx2 • t+1 • dx • B1 • B2 •  • t • dy • O • L • C2 • C1

  8. COMAD scheme (Malardel and Ricard, in review, QJ) 2 – Test of a modified Semi-Lagrangian scheme • * Computation of the trajectories: no modification • * Computation of the value of variables at the origin point  modification of the SL interpolation weights • For example, with linear interpolations (2D and regular grid): • Original SL scheme: • 2 linear zonal interpolations • VB = wx1 VB1 + wx2 VB2 with wx2 =  /dx, wx1 = 1 -  /dx • VC = wx1 VC1 + wx2 VC2 = 1 - wx2 • t+1 • dx • B1 • B2 • B •  • t • dy • O • L • C • C2 • C1

  9. COMAD scheme (Malardel and Ricard, in review, QJ) 2 – Test of a modified Semi-Lagrangian scheme • * Computation of the trajectories: no modification • * Computation of the value of variables at the origin point  modification of the SL interpolation weights • For example, with linear interpolations (2D and regular grid): • Original SL scheme: • 2 linear zonal interpolations • VB = wx1 VB1 + wx2 VB2 with wx2 =  /dx, wx1 = 1 -  /dx • VC = wx1 VC1 + wx2 VC2 = 1 - wx2 • 1 meridian linear interpolation • VO = wy1 VB + wy2 VC with wy1 = L / dy, wy2 = 1 - L /dy • t+1 • dx • B1 • B2 • B •  • t • dy • O • L • C • C2 • C1

  10. COMAD scheme (Malardel and Ricard, in review, QJ) 2 – Test of a modified Semi-Lagrangian scheme • * Computation of the trajectories: no modification • * Computation of the value of variables at the origin point  modification of the SL interpolation weights • For example, with linear interpolations (2D and regular grid): • COMAD scheme: • 2 linear zonal interpolations • VB = w’x1 VB1 + w’x2 VB2 with wx2 =  /dx, wx1 = 1 -  /dx • VC = w’x1 VC1 + w’x2 VC2 = 1 - wx2 • 1 meridian linear interpolation • VO =w’y1VB + w’y2 VC with wy1 = L / dy, wy2 = 1 - L /dy • t+1 • dx • B1 • B2 • B •  • t • dy • O • L • C • C2 • C1 • w’x1= xwx1 + 0.5 * (1- x) with x = (1 + U/ x * dt) deformation factor along x axis • w’x2 = xwx2 + 0.5 * (1- x) • w’y1 = ywy1 + 0.5 * (1- y) with y = (1 + U/ y * dt) deformation factor along y axis • w’y2= ywy2 + 0.5 * (1- y)

  11. COMAD scheme (Malardel and Ricard, in review, QJ) 2 – Test of a modified Semi-Lagrangian scheme • * Computation of the trajectories: no modification • * Computation of the value of variables at the origin point  modification of the SL interpolation weights • For example, with linear interpolations (2D and regular grid): • COMAD scheme: • 2 linear zonal interpolations • VB = w’x1 VB1 + w’x2 VB2 with wx2 =  /dx, wx1 = 1 -  /dx • VC = w’x1 VC1 + w’x2 VC2 = 1 - wx2 • 1 meridian linear interpolation • VO =w’y1VB + w’y2 VC with wy1 = L / dy, wy2 = 1 - L /dy • t+1 • A1 • A2 • B1 • B3 • B0 • B2 • t • O • C2 • C3 • C1 • C0 • D2 • D1 • w’x1= xwx1 + 0.5 * (1- x) with x = (1 + U/ x * dt) deformation factor along x axis • w’x2 = xwx2 + 0.5 * (1- x) • w’y1 = ywy1 + 0.5 * (1- y) with y = (1 + U/ y * dt) deformation factor along y axis • w’y2= ywy2 + 0.5 * (1- y) •  modified linear weights can also be used after for computing cubic weights

  12. COMAD scheme (Malardel and Ricard, in review, QJ) 2 – Test of a modified Semi-Lagrangian scheme • * Computation of the trajectories: no modification • * Computation of the value of variables at the origin point  modification of the SL interpolation weights • For example, with linear interpolations (2D and regular grid): • COMAD scheme: • 2 linear zonal interpolations • VB = w’x1 VB1 + w’x2 VB2 with wx2 =  /dx, wx1 = 1 -  /dx • VC = w’x1 VC1 + w’x2 VC2 = 1 - wx2 • 1 meridian linear interpolation • VO =w’y1VB + w’y2 VC with wy1 = L / dy, wy2 = 1 - L /dy • t+1 • A1 • A2 • B1 • B3 • B0 • B2 • t • O • C2 • C3 • C1 • C0 • D2 • D1 w’x1= xwx1 + 0.5 * (1- x) withx = (1 + U/ x * dt) deformation factor along x axis w’x2 = xwx2 + 0.5 * (1- x) w’y1 = ywy1 + 0.5 * (1- y) withy = (1 + U/ y * dt) deformation factor along y axis w’y2= ywy2 + 0.5 * (1- y) • modifiedlinearweightscanalsobeusedafter for computingcubicweights AROME uses quasi-cubic interpolations (2 linear, 3 cubicones)

  13. Example: 30 June 2012 2 – Test of a modified Semi-Lagrangian scheme • OPER SL • COMAD • 24-h precipitation (mm) from 00 UTC - Wind vectors at 10 m (m/s), 00 UTC 1 July • Less precipitation • Less intense wind ahead of precipitation area

  14. Example: 30 June 2012 2 – Test of a modified Semi-Lagrangian scheme • OPER SL • COMAD • 3-h precipitation (mm) 15-18 UTC, Wind vectors at 10 m (m/s) 18 UTC 30 June • Less intense convective cells • Less intense outflows

  15. Example: 30 June 2012 2 – Test of a modified Semi-Lagrangian scheme • OPER SL • COMAD • 3-h precipitation (mm) 15-18 UTC, Wind vectors at 10 m (m/s) 18 UTC 30 June • Less intense convective cells • Less intense outflows

  16. Example: 30 June 2012 2 – Test of a modified Semi-Lagrangian scheme • OPER SL • COMAD • Virtual potential temperature (K) - Wind vectors at 10 m (m/s), 18 UTC 30 June • Less intense convective cells • Less intense cold pools

  17. 15 July - 15 September 2013 2 – Test of a modified Semi-Lagrangian scheme • OPER SL • COMAD • Mean 24-h precipitation over the forecast domain • Less precipitation amount

  18. 15 July - 15 September 2013 2 – Test of a modified Semi-Lagrangian scheme • Mean 24-h precipitation over the forecast domain • Less precipitation amount • Variation between 1 and –26 %

  19. 15 July - 15 September 2013 2 – Test of a modified Semi-Lagrangian scheme • OPER SL • COMAD • 24-h precipitation distribution for all gridpoints of the forecast domain • Smaller frequencies of moderate and heavy precipitation

  20. Scores:15 July - 15 September 2013 2 – Test of a modified Semi-Lagrangian scheme • OPER SL • mse • COMAD • bias • 6-h precipitation (mm) • Forecast range (hour) • mse • Surface pressure (hPa) • bias • Forecast range (hour) • 6-h precipitation: better scores • Surface pressure: slight improvement for bias

  21. Scores:15 July - 15 September 2013 2 – Test of a modified Semi-Lagrangian scheme • OPER SL • mse • COMAD • 2m temperature (K) • bias • Forecast range (hour) • mse • 10m Wind intensity (m/s) • bias • Forecast range (hour) • Near-surface wind and temperature: slight degradation after 18h forecast

  22. Fuzzy scores: 15 July - 15 September 2013 2 – Test of a modified Semi-Lagrangian scheme • OPER SL • RR24 > 5 mm • RR24 > 0.2mm • COMAD • Neighbourhood (km) • Neighbourhood (km) • RR24 > 10 mm • RR24 > 20mm • Neighbourhood (km) • Neighbourhood (km) • Brier Skill Scores for 24-h precipitation (06UTC-06UTC) • Better scores for all thresholds and all neighbourhoods

  23. Fuzzy scores: 15 July - 15 September 2013 2 – Test of a modified Semi-Lagrangian scheme • OPER SL • RR6 > 2 mm • RR6 > 0.5 mm • COMAD • Neighbourhood (km) • Neighbourhood (km) • RR6 > 5 mm • RR6 > 10mm • Neighbourhood (km) • Neighbourhood (km) • Brier Skill Scores for 6-h precipitation (12UTC-18UTC) • Better scores for all thresholds and all neighbourhoods

  24. Example: 30 June 2012 2 – Test of a modified Semi-Lagrangian scheme • OPER SL • COMAD • Running variance (100 km * 100 km) of wind at 10 m (m/s)², 18 UTC 30 June • Less intense convective cells • Less intense downdrafts

  25. 15 July -15 September 2013 2 – Test of a modified Semi-Lagrangian scheme • 10-m downdrafts (m²/s²) • 10-m Wind (m²/s²) • OPER SL • COMAD • Running variance (100 km * 100 km) • (hourly averaged over the forecast domain and the period 15 July - 15 September 2013) • Less variance during the afternoon and evening • Less intense density currents under convective cells • 925 hPa Virtual potential temperature (K²)

  26. 15 July -15 September 2013 2 – Test of a modified Semi-Lagrangian scheme • OPER SL • COMAD • Diurnal cycle of surface covered by convective cells (simulated reflectivities above 30 dBZ) • Less intense convective cells

  27. Methodology 3 – Evaluation of AROME at kilometric resolution • L 90BC • L 90 • L 60 • Smaller forecast domain • (720 points *720 points - 1.3km) • (360 points *360 points - 2.5km) • Layer thickness (m) Configuration: • for stability: ICI scheme (instead of 2TL SI scheme) • time step: 45s (instead of 60s) • initial conditions: dynamical adaptation from 2.5km 3DVAR Analysis • LBC: from operational AROME • better representation of the orography at 1.3km

  28. Methodology 3 – Evaluation of AROME at kilometric resolution Period: 1 June-30 November 2012 Selection of days with moderate and intense convective activity over the forecast domain • lightning data (more than 5000 strikes per day) • 48 days • 24-h lightning data (21 June) : 88897 lightning strikes

  29. Scores 3 – Evaluation of AROME at kilometric resolution • Classic scores (bias, MSE) • Fuzzy scores (Brier Skill scores) • Increase of vertical resolution: • better classic scores (temp and humidity) but no better fuzzy scores

  30. Scores 3 – Evaluation of AROME at kilometric resolution • Classic scores (bias, MSE) • Fuzzy scores (Brier Skill scores) • Increase of vertical resolution: • better classic scores (temp and humidity) but no better fuzzy scores • Increase of horizontal resolution: • better fuzzy scores • degradation for temperature and humidity scores but improvement for wind score

  31. Scores 3 – Evaluation of AROME at kilometric resolution • Classic scores (bias, MSE) • Fuzzy scores (Brier Skill scores) • Increase of vertical resolution: • better classic scores (temp and humidity) but no better fuzzy scores • Increase of horizontal resolution: • better fuzzy scores • degradation for temperature and humidity scores but improvement for wind score • No further improvement with more levels below 2000m

  32. Characteristics of convective cells • dbZ • 5 • 10 • 15 • 20 • 30 • 40 • 50 3 – Evaluation of AROME at kilometric resolution Comparison to observations using a tracking algorithm (Morel et al., 2002) to detect convective cells (2 thresholds > 30 dBZ and > 40 dBZ)  size, number, intensity maximum of convective cells • 2.5km: 76 cells > 40 dBZ • Simulated reflectivities at 1500 m 21 June 12UTC

  33. Characteristics of convective cells • dbZ • 5 • 10 • 15 • 20 • 30 • 40 • 50 3 – Evaluation of AROME at kilometric resolution Comparison to observations using a tracking algorithm (Morel et al., 2002) to detect convective cells (2 thresholds > 30 dBZ and > 40 dBZ)  size, number, intensity maximum of convective cells • 2.5km: 76 cells > 40 dBZ • Simulated reflectivities at 1500 m 21 June 12UTC

  34. Characteristics of convective cells • dbZ • 5 • 10 • 15 • 20 • 30 • 40 • 50 3 – Evaluation of AROME at kilometric resolution Comparison to observations using a tracking algorithm (Morel et al., 2002) to detect convective cells (2 thresholds > 30 dBZ and > 40 dBZ)  size, number, intensity maximum of convective cells • 2.5km: 76 cells > 40 dBZ • 1.3km: 122 cells > 40 dBZ • Simulated reflectivities at 1500 m 21 June 12UTC

  35. Characteristics of convective cells > 40 dBZ - 21 June 3 – Evaluation of AROME at kilometric resolution • 1.3km • 1.3km • radar • radar • 2.5km • 2.5km • Time evolution of cell number • Surface distribution • 1.3 km vs 2.5km: • more cells • more numerous small cells • fewer large cells • more realistic

  36. Characteristics of convective cells > 30dBZ and > 40 dBZ - 48 days 3 – Evaluation of AROME at kilometric resolution • radar • 1.3km • radar • 1.3km • 2.5km • 2.5km • Surface distribution > 40dBZ • Surface distribution > 30dBZ • Over the 48 days at the peak of convection, 1.3 km vs 2.5km: • more realistic • more numerous small and medium cells • fewer large cells

  37. Conclusion • Increase of horizontal grid spacing (1.3km versus 2.5km): • more realistic number of cells • more numerous small cells, fewer large cells • reduction of precipitation amount • better fuzzy scores (for precipitation, brightness temperature, downdrafts …) • Use of the modified SL scheme (COMAD versus original SL scheme) • less intense convective cells • improvement of QPF, less amount • better fuzzy scores for precipitation •  test on other periods: June 2012, January 2013 (frontal precipitation) •  Test of the modified SL scheme at 1.3km

  38. Fuzzy scores: 15 July - 15 September 2013 2 – Test of a modified Semi-Lagrangian scheme • OPER SL • Neighbourhood 52 km • Neighbourhood 20 km • COMAD • Temperature thresholds (K) • Temperature thresholds (K) • Brier Skill Scores for brightness temperature 10.8 m (forecast range 18 UTC) • For peak of convection: better scores in particular for lower temperature thresholds •  better representation of the high clouds

  39. 1-31 January 2013 2 – Test of a modified Semi-Lagrangian scheme • OPER SL • COMAD • Mean 24-h precipitation over the forecast domain • Less impact on frontal precipitation

  40. 1-31 January 2013 2 – Test of a modified Semi-Lagrangian scheme • Mean 24-h precipitation over the forecast domain • Less impact on frontal precipitation • Variation between 1 and –5 %

  41. Fuzzy scores: 1-31 January 2013 2 – Test of a modified Semi-Lagrangian scheme • OPER SL • RR24 > 5 mm • RR24 > 0.2mm • COMAD • Neighbourhood (km) • Neighbourhood (km) • RR24 > 10 mm • RR24 > 20mm • Neighbourhood (km) • Neighbourhood (km) • Brier Skill Scores for 24-h precipitation (06UTC-06UTC)

  42. 1-30 June 2012 2 – Test of a modified Semi-Lagrangian scheme • OPER • MODIFSL • Mean 24-h precipitation over the forecast domain • Less precipitation amount

  43. 1-30 June 2012 2 – Test of a modified Semi-Lagrangian scheme • Mean 24-h precipitation over the forecast domain • Less precipitation amount • Reduction between –1 and –25 %

  44. 1-30 June 2012 2 – Test of a modified Semi-Lagrangian scheme • COMAD • OPER SL • 24-h precipitation distribution for all gridpoints of the forecast domain • Smaller frequencies of moderate and heavy precipitation

  45. Fuzzy scores: 1-30 June 2012 2 – Test of a modified Semi-Lagrangian scheme • RR24 > 1 mm • RR24 > 0.2mm • MODIFSL • OPER • Neighbourhood (km) • Neighbourhood (km) • RR24 > 10 mm • RR24 > 20mm • Neighbourhood (km) • Neighbourhood (km) • Brier Skill Scores for 24-h precipitation (forecast range 30h) • Better scores for all thresholds and all neighbourhoods

  46. Fuzzy scores: 1-30 June 2012 2 – Test of a modified Semi-Lagrangian scheme • MODIFSL • Neighbourhood 120 km • Neighbourhood 20 km • OPER • Temperature thresholds (K) • Temperature thresholds (K) • Brier Skill Scores for brightness temperature 10.8 m (forecast range 18 UTC) • For peak of convection: better scores in particular for lower temperature thresholds •  better representation of the high clouds

  47. Example: 30 June 2012 2 – Test of a modified Semi-Lagrangian scheme • OPER SL • COMAD • Running variance (100 km * 100 km) of wind at 10 m (m/s)², 18 UTC 30 June • Less intense convective cells • Less intense downdrafts

  48. Example: 30 June 2012 2 – Test of a modified Semi-Lagrangian scheme • OPER SL • COMAD • Running variance (100 km * 100 km) of downdrafts at 10 m (m/s)², 18 UTC 30 June • Less intense convective cells • Less intense downdrafts

  49. Example: 30 June 2012 2 – Test of a modified Semi-Lagrangian scheme • OPER SL • COMAD • Running variance (100 km * 100 km) of 925 hPa ϴv at 10 m (K)², 18 UTC 30 June • Less intense convective cells • Less intense downdrafts

More Related