Скачать презентацию Status report of WG 2 Numerical aspects
29d80ae54ad1a145dbd8239c7fc3c239.ppt
- Количество слайдов: 52
Status report of WG 2 – Numerical aspects COSMO General meeting, Israel, Jerusalem 11 -14 Sept. 2017 Michael Baldauf, Uli Blahak (DWD), Guy de Morsier, Pascal Spörri (Meteo. CH), Andreas Will, Jack Ogaja (Univ. Cottbus), Werner Schneider (Univ. Bonn)
Outlook • • Investigations for the new COSMO-D 2 setup The new Bott (2010) advection scheme Higher order scheme (horizontal) Extended targeted diffusion against cold pools • PP CDIC • PP CELO Bogdan Rosa • new PP EX-CELO Zbigniew Piotrowski M. Baldauf (DWD) 2
At DWD, COSMO-D 2 will replace the current COSMO-DE with the following changes: è increase horizontal resolution from 2. 8 km to 2. 2 km è increase number of vertical levels from 50 to 65 è increase area from 10. 5° * 11. 5° to 13° * 14. 3° COSMO-D 2: 651 * 716 * 65 GPe 1440 * 1590 * 22 km³ COSMO-DE: 421 * 461 * 50 GPe 1160 * 1280 * 22 km³ Time schedule: since June 2017: pre-operational phase Q 2/2018: operational introduction M. Baldauf (DWD) 3
New choice of vertical levels in COSMO-D 2: • increase resolution mainly of the boundary layer goal: improve initiation of convection • change from Gal-Chen to SLEVE-coordinate COSMO-D 2 (SLEVE) COSMO-DE B. Ritter, B. Fay, U. Schättler, A. Seifert, C. Schraff, H. Reich, C. Gebhardt, M. Buchhold, F. Fundel, T. Hanisch, H. Frank, … M. Baldauf (DWD) 4
Frequency spectra for
Power spectra of kinetic energy KE spectra COSMO-D 2 COSMO-DE SW-inflow, some heavy showers M. Baldauf, B. Ritter (DWD) 6
KE spectra COSMO-D 2 COSMO-DE M. Baldauf, B. Ritter (DWD) 7
KE spectra COSMO-D 2 COSMO-DE M. Baldauf, B. Ritter (DWD) 8
Power spectra of vertical velocity w² spectra COSMO-D 2 COSMO-DE SW-inflow, some heavy showers M. Baldauf (DWD) 9
Conclusions from powerspectra • nothing goes obviously wrong with the 2. 2 km resolution: spectra for C-D 2 are relatively similar to those of C-DE (C-DE KE spectrum at 12. 08. 17 + 12 h has a strange increase at small scales) • effective or 'practical' resolution: • often for t=0 (analysis time) similar for C-D 2 and C-DE • for later forecast times mostly (but not always!) higher for C-D 2 than for C-DE (as it should be) • only in a few cases a k -3 -behaviour is visible on the large scale (probably the area is too small) • sometimes a (weak) maximum in w² spectrum around 4 dx is visible. Skamarock et al. (2014) JAS: "We believe that the filter-scale peak is likely associated with grid-scale convection, waves generated by convection, and other marginally or underresolved small-scale processes. " M. Baldauf (DWD) 10
The new Bott advection scheme … as an optional candidate for tracer advection Bott (2010) MWR: Werner Schneider (Univ Bonn) Uli Blahak (DWD) • tries to solve the 'mass-consistency'-problem without parallel computation of a continuity equation, but with an add. /subtr. of the divergence in the direction-splitting scheme • without full Strang-splitting efficiency gain; however still x-y-z / z-y-x for odd/even time steps • 4 th order scheme Following slides: Verification results for the comparison of operational COSMO-DE and COSMO-DE with new Bott scheme M. Baldauf (DWD) 11
Synop verification Bott (2010) current M. Baldauf (DWD) 12
Synop verification Bott (2010) current M. Baldauf (DWD) 14
Upper air verification M. Baldauf (DWD) 15
Summary for the verification results of the new Bott-scheme • Synop-Verif. of T 2 m and v 10 m is slightly positive, neutral for TD 2 m, RH 2 m • Synop-Verif. of categorical measures for rain and gusts is negative, cloudiness is neutral • Temp-Verif. is very positive for the new Bott-scheme Proposal: the results are not entirely satisfying, however good enough to bring the new Bott-scheme as an option (!) into the official code (v 5. 6) (fulfil COSMO science plan , sec. 5. 2. 4) M. Baldauf (DWD) 17
Higher order discretization Status A. Will, J. Ogaja (BTU Cottbus) • Large improvement in efficiency done! In the comparison adv. S 4 -P 4 / adv. UP 5 -P 2: advection ~10%, fast waves ~3% more expensive. without artificial diffusion, the costs are roughly the same! • Dissertation J. Ogaja is available. • Model crash COSMO-DE at ‚ 20. 06. 2013‘ has been solved • Summerly precipitation dry bias is much reduced in the convection-permitting setup! Next steps: • Transfer the code from version 5. 0 to 5. 6 ( A. Will, M. Baldauf) • Run this new version on NUMEX ( M. Baldauf) • Deliver documentation (COSMO Sci. Doc. part I, possibly also a COSMO-TR (? )) ( A. Will) • expected availability for the official code version: ~June 2017 ~Dec. 2017 M. Baldauf (DWD) 18
Extended targeted diffusion against cold pools G. de Morsier, P. Spörri (Meteo. CH) Motivation: 5 th order advection can produce artificially cold pools in narrow valleys targeted diffusion necessary. However, the current implementation (diffusion with 5 -point stencil, only applied for T) did not cure every cold pool ocuring in COSMO-1 (1. 1 km) or COSMO-E (2. 2 km, 21 members, with SPPT); (both setups don't use any artificial horizontal diffusion in inner domain!) Proposal: extend targeted diffusion to a 9 -point stencil and apply to T, u, v where needed. This cured every T anomaly in a 6 month experiment! Status: Code ready for COSMO v 5. 6 (only a switch still necessary) for both Fortran and STELLA code version. M. Baldauf (DWD) 19
Temperature Anomaly Problem (2) © Guy de Morsier, 11 Sept. 2017, Jerusalem 20
Temperature Anomaly © Guy de Morsier, 11 Sept. 2017, Jerusalem 21
Status report of the Priority Project ‚Comparison between the dynamical cores of COSMO and ICON‘ (CDIC) COSMO General meeting, Jerusalem 11 -14 September 2017 Project team (current): Michael Baldauf, Florian Prill (DWD), Rodica Dumitrache, Amalia Iriza (NMA), Damian Wojcik (IMGW), Guy de Morsier (Meteo. CH), Marina Shatunova, Denis Blinov, Alexandr Kirsanov (Roshydromet) with strong support from Günther Zängl, Daniel Reinert, Uli Schättler (DWD) M. Baldauf (DWD) 22
Aim of the COSMO priority project ‚Comparison of the dynamical cores of COSMO and ICON‘ (CDIC): deliver an as objective as possible comparison of the dynamical cores of COSMO and ICON with the emphasis on limited area modelling. • • • Task 1: idealised tests (main focus) Task 2: semi-realistic tests Task 3: scalability/performance Task 4: Principal properties of the numerical formulation Task 5: Suitability for other applications (climate/chemistry) M. Baldauf (DWD) 23
Task 1. Good performance on a standard set of idealized test cases Defined test cases 1. Advection test with nonlinear dynamics (Schär et al. , 2002) ? 2. Atmosphere at rest (Zängl et. al (2004) Met. Z) 3. Cold bubble unstationary density flow (Straka et al. , 1993) 4. Mountain flow tests (stationary, orographic flows) 4. 1 Schär et al. (2002), section 5 b 4. 2 Bonaventura (2000) JCP (selection) 4. 3 3 D-case (dry) 5. Linear Gravity waves (Baldauf, Brdar, 2013) 6. Warm bubble (Robert (1993), Giraldo (2008)) 7. Moist, warm bubble (Weisman, Klemp, 1982) 8. Advection tests for tracer schemes (solid body rotation, …) ! M. Baldauf (DWD) 24
Test case 4. 1: 2 D linear flow over mountains COSMO x=500 m z=300 m colors and black dotted lines: COSMO or ICON grey lines: analytic solution (Baldauf, 2008, COSMO-NL) ICON vertically equidistant grid M. Baldauf (DWD) 27
Test case 4. 3 a: 3 D linear flow over mountains COSMO ICON colors and grey lines : COSMO or ICON simulation black lines: analytic solution (Baldauf, 2008, COSMO-NL) M. Baldauf (DWD) 31
Test case 4. 3 a: 3 D linear flow over mountains COSMO ICON colors and grey lines : COSMO or ICON simulation black lines: analytic solution M. Baldauf (DWD) 32
Test case 4. 3 b: 3 D nonlinear flow over mountains COSMO ICON h 0=1000 m max h = 234. 9 m max = 13. 2° M. Baldauf (DWD) 34
Test case 4. 3 b: 3 D nonlinear flow over mountains COSMO ICON h 0=3000 m max h = 704. 7 m max = 35. 2° M. Baldauf (DWD) 35
Test case 4. 3 b: 3 D nonlinear flow over mountains COSMO ICON stable only with Mahrer-discretization h 0=4000 m max h = 939. 6 m max = 43. 2° M. Baldauf (DWD) 36
Test case 4. 3 b: 3 D nonlinear flow over mountains COSMO ICON COSMO: unstable until h 0=4600 m max h = 1080 m max = 47. 2° h 0=5000 m max h = 1174 m max = 49. 6° M. Baldauf (DWD) 37
Test case 4. 3 b: 3 D nonlinear flow over mountains COSMO ICON COSMO: unstable h 0=8000 m max h = 1879 m max = 62. 0° M. Baldauf (DWD) 38
Summary for test cases 4. x: flow over mountains tests • In the Schär et al. 2 D linear mountain flow test both models COSMO and ICON behave quite similar; with slight advantages for ICON. • Also in the 3 D linear test the analytic solution is very well simulated metric terms are correctly implemented in both models (no clear winner) • ICON tolerates much steeper slopes than COSMO (Zängl (2012) MWR) • The high mountain tests should be repeated with ‚non-periodic BCs‘ to prevent from increasing blocking effects M. Baldauf (DWD) 39
Test case 5: linear gravity + sound waves setup similar to Skamarock, Klemp (1994) MWR An analytic solution for the compressible non-hydrostatic Euler equations is given in Baldauf, Brdar (2013) QJRMS Test properties: • test dry Euler equations • unstationary inspect time integr. • no orography • small amplitude linear comparison with analytic solution M. Baldauf (DWD) 40
Test case 5: linear gravity + sound waves Convergence behaviour COSMO ICON T‘ r d 2 n e ord 1 st order w due to a bug fix in the test setup (proper use of periodic BCs) the COSMO result is now better than that described in BB 2013 M. Baldauf (DWD) 42
Summary for test case 5: linear gravity + sound waves • Test 1 (only fast waves): ICON shows nearly 2 nd order convergence COSMO shows nearly 2 nd order only in T, but less in w w error is smaller in ICON for fine resolutions • Test 2 (FW + advection): ICON behaviour is similar to test 1. COSMO convergence order is slightly reduced for coarse resolutions ICON errors are a bit larger than in COSMO, for fine resolutions a bit smaller • Test 3 (FW + Coriolis): both models show 2 nd order convergence; but the errors are smaller in ICON Remark: to get 2 nd order, one needs to switch off any vertical off-centering M. Baldauf (DWD) 45
Test case 3: cold bubble R. Dumitrache, A. Iriza (NMA) Testsetup by Straka et al (1993) Test properties: • test of dry Euler equations (without Coriolis force) • unstationary • strongly nonlinear • comparison with reference solution from paper M. Baldauf (DWD) 46
COSMO Test case 3: cold bubble at t=15 min. for x= 200, 100, 50, 25 m ICON something goes wrong… diffusion? still to be done… Reference solution from Straka et al. : M. Baldauf (DWD) 48
Test case 2: atmosphere at rest M. Baldauf (DWD) Test setup similar to Zängl et al. (2004) same settings in COSMO and ICON for • grid (COSMO quadrilat. , ICON triangle with dx=2 km, dz=19. 8 m… 780 m, htop=20 km, vcflat=15 km, rdheight=16 km) • reference atmosphere (irefatm=2) • initial profile: piecewise polytropic M. Baldauf (DWD) 50
Test case 2: atmosphere at rest ICON COSMO y_vert_adv_dyn='impl 2' y_vert_adv_dyn='impl 3' M. Baldauf (DWD) 51
Test case 2: atmosphere at rest ICON COSMO y_vert_adv_dyn='impl 2' y_vert_adv_dyn='impl 3' M. Baldauf (DWD) 52
Test case 2: atmosphere at rest ICON COSMO y_vert_adv_dyn='impl 2' y_vert_adv_dyn='impl 3' M. Baldauf (DWD) 53
Test case 2: atmosphere at rest time series of wmax COSMO, impl 2 ICON COSMO, impl 3 COSMO, expl M. Baldauf (DWD) 54
Summary for test case 2: atmosphere at rest • disturbances sets in faster in COSMO than in ICON • the maximum vertical velocity at saturation is smaller in COSMO (at least for first 24 h) • In COSMO the operational (!) setting y_vert_adv_dyn='impl 2' is unstable, only 'expl' and 'impl 3' remain stable. However, the latter two options both need additional investigations. M. Baldauf (DWD) 55
Test case 7: Weisman, Klemp (1982) test setup D. Wojcik (IMGW) • Idealized moist convection experiment designed to reproduce the development and subsequent evolution of a convective cloud • Test basic consistency of the coupling between dry model equations and moist microphysics and turbulence parameterization horizontal resolution: 2 km COSMO, ICON: microphysics with 3 -cat. ice WK 82: Kessler Following plots show: • Vertical wind: on 4150 m height (contours, negative values dashed) • Horizontal wind: on 87 m height (arrows) • Gust front: on 10 m height (thick blue line, - 0. 5 K temperature perturbation) • Precipitation: on 10 m height (dashed, for QR values exceeding 1 and 4 g / kg) M. Baldauf (DWD) 56
Results for Us = 25 m/s ICON model COSMO model 40 min 80 min WG 2 COSMO General Meeting 2017 W-K 1982
Results for Us = 25 m/s ICON model COSMO model 120 min WG 2 COSMO General Meeting 2017 W-K 1982
Summary 1. In this experiment ICON model demonstrates capability to reproduce realistic convective nonhydrostatic flows 2. There is no indication of basic errors in the coupling between dry ICON dynamical core and moist microphysics and turbulence parameterizations 3. When rather little horizontal environmental vorticity in present (Us = 15 m/s) the ICON model reproduces basic convective structures (gust front, convective updraft, precipitation region and surface outflow). The convective updraft tends to get more ‘blurred’, but that probably results from different effective diffusivity of the two models 4. In the middle case (Us = 25 m/s) the convective updraft for the ICON model is more compact (comparing to Us = 15 m/s) bus still less compact in comparison with COSMO R-K 5. In the case with high env. vorticity (Us = 35 m/s) the ICON model updraft is more compact comparing with COSMO R-K and more similar to the benchmark solution. Also the lateral drift of the convective cell is more similar to the benchmark solution WG 2 COSMO General Meeting 2017
Overall conclusions for task 1 • Most of the planned idealised tests have been inspected, almost all of these have been finished • In general, no detrimental effects of ICON visible (in the contrary!) • However, the question remains „what is a fair comparison“? E. g. quadrilateral vs. triangle grid, …: what is the ‚right‘ resolution? Probably the best is to compare „error as fct. of model run time“ (on the same computer) (but this needs some extra considerations for 2 D slice tests) M. Baldauf (DWD) 61
Task 2: Ability to handle semi-realistic cases reasonably well Test cases are defined: • strong advective case: storm ‚Elon‘, 9 -10 Jan. 2015 (Meteo. CH) • strong advective case: storm 'Carmen', 12 Nov. 2010 (Meteo. CH) • Bora event: 6 -8 Feb. 2012 (possibly also 19 Feb. 2016) (RHM) However, the ICON developers didn‘t want to distribute the ‚limited area mode‘ version before the ICON trainings course 28. 02. -03. 17 because documentation will not be ready before this event. heavy delay in this task project prolongation required M. Baldauf (DWD) 62
Task 3. Scalability/Performance suitable for operations as well as for future supercomputing platforms strong scaling tests on Cray XC 40 (Broadwell): COSMO-D 2 and ICON-D 2 setups, 27 h forecast runs Code improvements in ICON for reading of boundary data have been necessary before starting the strong scalability tests (F. Prill) Further optimisation planned by reducing latency by collecting all levels before communication. Comparison: Efficiency of ICON is higher than in COSMO. However, the apparently reduced scalability in ICON for #procs>1000 still stems from the distribution of boundary data (even influences time measurement of dynamics) M. Baldauf (DWD) 63
Task 3 COSMO-D 2 this bad scaling (caused by RTTOV) has been improved by U. Schättler in COSMO 5. 4 f probably needed for COSMO-D 2 M. Baldauf (DWD) 64
Task 3 ICON-D 2 no output Boundary data exchange must be improved todo: separation of comput. and communication in the ICON time measurement of dynamics and physics, M. Baldauf (DWD) 65
Task 4: Identification of differences in dynamical core formulations stability analysis of both dynamical cores has been done Task 5. Suitability of ICON dynamical core for other applications than NWP (climate, chemistry, . . . ) compared to the COSMO model assessment of: ICON-ART (Roshydromet), climate version (AG ICON at CCLM) Heavy delay for the same reasons as in task 2 COSMO STC has approved the prolongation of PP CDIC until Aug. 2018 ! M. Baldauf (DWD) 66