Modeling of regional atmospheric circulation using high-resolution hydrodynamical model Mikhail A. Tolstykh, Institute of Numerical Mathematics, RAS, and Hydrometcentre of Russia
Outline • SLAV model – what it is? • Recent developments and results, in particular for Asia and Western Siberia • Towards high-resolution global model: necessary developments
SLAV model • Global finite-difference semi-Lagrangian semi-implicit dynamical core of own development + parameterizations of subgrid-scale processes from French model ARPEGE/ALADIN. • Distinct features of dynamical core – vorticity-divergence formulation on the unstaggered grid, wide use of 4 th order finite differences, usual and compact. High accuracy of dynamical core ( Tolstykh, J. Comput. Phys. 2002). • Computational efficiency due to semi-Lagrangian advection and good parallel implementation.
Perform ance and scalability of testSLM on p575 DC 16 CPUs@ 1.5G Hz Параллельная реализация перспективной tim e (s ec) версии на IBM 575 10000.0 160.00 146.67 (OpenMP + MPI, только динамический блок ) 9732 Speed-up 140.00 C ode tim e (sec/step) T im e com m and (sec) 1000.0 120.00 352.0 741 100.00 92.63 450 306 238 189 100.0 80.00 22.6 53.33 60.00 11.8 6.6 10.0 40.00 29.83 3.8 2.4 15.58 20.00 1.00 1.0 0.00 1 mpi * 1 omp on 8mpi * 4omp on 1 6mpi * 4omp on 32mpi * 4omp on 64mpi * 4omp on 64mpi * 8omp on 1 6CPUs w ith 1 6CPUs w ith 32CPUs w ith 64CPUs w ith 1 28CPUs w ith 256CPUs w ith SMT=NO SMT=YES SMT=YES SMT=YES SMT=YES SMT=YES
Current state of the constant resolution global semi-Lagrangian model SL-AV • Horizontal resolution 0,9 ˚ x0,72 ˚ lon-lat, 28 vertical levels. • Implemented operationally. • In particular, forecasts for Tomsk area are available via ftp for TSU • Initial data – RHMC operational analyses + T2m and RH2m own OI analysis, surface pressure field from RHMC DAS • Improved annual mean scores in 2006 vs 2005, especially in NH and Asia. • One-year operational tests for precipitation forecasts finished 30 th of June. • Operational tests for T2m forecasts started on 1 st of July 2007.
Averaged RMS error for MSLP forecasts. Start time: 12 UTC. 27/10-23/12 2006 . Region: Northern extratropical hemishpere 8,00 7,00 6,00 5,00 RMSE, mb 4,00 SLMop 3,00 2,00 SLMex 1,00 0,00 0 24 48 72 96 120 Forecast range (hours)
Innovation vectors (dots), analysis increments (color field) for T2m RHMC DAS analysis
Variable resolution version of SLAV model • Horizontal resolution 0,5625 ˚ in longitude, 26-70 in latitude, 28 vertical levels. • Initial data – interpolated initial data for constant resolution version
Resolution in latitude as a function of latitude (in degrees)
Recent developments • Implementation of finite-element scheme for integration of hydrostatics equation (see poster by A.V.Shlyaeva, M.A.Tolstykh) • Development of a nonhydrostatic dynamical core (see poster by R.Yu. Fadeev, M.A.Tolstykh) • Assimilation of soil variables; T2m and RH2m OI analysis, development and first results for 3D-Var assimilation for T2m field (see poster by N.N.Bogoslovskii, A.V.Shlyaeva, M.A.Tolstykh)
Forecast quality of upper air fields over Asia in 2006 Models compared: - ECMWF – European centre for medium range weather forecasts - UKMO – UK Meteorological office - NCEP – USA MRF model - SMA – Russian T85L31 Eulerian model - SLAV – Russian semi-Lagrangian model, constant resolution version - SLAV-VR – variable resolution version of SLAV model
RMS error of H500 forecast for Asia. Start time 12 UTC. Period: 2006. Verification against analyses 14,0 12,0 10,0 RMSE, dam 8,0 SMA 6,0 ECMWF UKMO 4,0 NCEP SLAV 2,0 SLAV VR 0,0 0 24 48 72 96 120 144 168 192 216 240 Forecast range (hours)
RMS error of MSLP forecast for Asia. Start time: 12 UTC. Period: 2006. Verification against analyses 16,0 14,0 12,0 10,0 RMSE, hPa 8,0 SMA 6,0 ECMWF 4,0 UKMO 2,0 SLAV SLAV VR 0,0 0 24 48 72 96 120 144 168 192 216 240 Forecast range (hours)
RMS error of T850 forecast for Asia. Start time: 12 UTC. Period: 2006. Verification against analyses 8,0 7,0 6,0 5,0 RMSE, о С SMA 4,0 ECMWF 3,0 UKMO NCEP 2,0 SLAV 1,0 SLAV VR 0,0 0 24 48 72 96 120 144 168 192 216 240 Forecast range
RMS error of V250 for Asia. Start time 12 UTC. Period: 2006 Verification against analyses 25,0 20,0 15,0 RMSE, m/s 10,0 SMA UKMO SLAV 5,0 SLAV VR 0,0 0 24 48 72 96 120 144 168 192 216 240 Forecast range (hours)
Verification of precipitation forecasts • Region – 50-61 ˚ N, 60-85 ˚ E. • Occurrence of precipitation: proportion correct, Heidke skill score, Pearcy criteria. • Quantiative precipitation forecast - absolute error for following grades of forecast: rain (0,3- 1,0; 1,1-10,5; 10,6-19,5 mm/12 h) ; snow (0,2-1,0; 1,1-4,5; 4,6-9,5 mm/12 h). • Separately for summer period (July-September 2006, May-June 2007) and winter period (October 2006 – April 2007)
Models compared: • EXE – UK MO global model • NCEP – USA MRF model • T85 – Russian Eulerian T85L31 model • Reg – “regional” model with 75 km resolution • SLAV – constant resolution semi- Lagrangian model • SLAV-VR – variable-resolution version of SLAV
Proportion correct for precipitation forecasts EXE NCEP over Siberia (%). October 2006 - April 2007. T85 Reg 90,0 SLM SLMVar 80,0 70,0 60,0 50,0 40,0 30,0 20,0 10,0 0,0 12 24 36 48 60 72 Forecast range (hours)
EXE Heidke skill score (HSS) x 100. NCEP October 2006 - April 2007. T85 60,0 Reg SLM 50,0 SLMVar 40,0 30,0 20,0 10,0 0,0 12 24 36 48 60 72 Forecast range (hours)
Pearcy criteria х 100. EXE October 2006 - April 2007 NCEP T85 70,0 Reg 60,0 SLM SLMVar 50,0 40,0 30,0 20,0 10,0 0,0 12 24 36 48 60 72 Forecast range (hours)
Abs error (mm) of snow forecast for range 0,2-1,0 mm/12 h. Ocotber 2006 - March 2007. EXE 2,5 NCEP T85 Reg 2,0 SLM SLMVar 1,5 1,0 0,5 0,0 12 24 36 48 60 72 Forecast range (hours)
Abs error (mm) of snow forecast for range EXE 1,1-4,5 mm/12 h. NCEP Ocotber 2006 - March 2007. T85 4,0 Reg 3,5 SLM SLMVar 3,0 2,5 2,0 1,5 1,0 0,5 0,0 12 24 36 48 60 72 Forecast range (hours)
Abs error (mm) of snow forecast for EXE range 4,6 - 9,5 mm/12 h. NCEP October 2006 - March 2007 T85 6,0 Reg SLM 5,0 SLMVar 4,0 3,0 2,0 1,0 0,0 12 24 36 48 60 72 Forecast range (hours)
EXE NCEP Precipitation forecast - proportion correct (%) T85 July-September 2006. Reg 90,0 SLM 80,0 SLMVar 70,0 60,0 50,0 40,0 30,0 20,0 10,0 0,0 12 24 36 48 60 72 Forecast range (hours)
Heidke skil score (HSS) x 100 EXE July-September 2006. NCEP 60,0 T85 Reg 50,0 SLM SLMVar 40,0 30,0 20,0 10,0 0,0 12 24 36 48 60 72 Forecast range (hours)
Pearcy criteria x 100 July-September 2006 . EXE NCEP 60,0 T85 Reg 50,0 SLM SLMVar 40,0 30,0 20,0 10,0 0,0 12 24 36 48 60 72 Forecast range (hours)
Abs. Error of precipitation forecast for EXE grade 0,3-1,0 mm/12h. NCEP July -September 2006. T85 3,00 Reg SLM 2,50 SLMVar 2,00 1,50 1,00 0,50 0,00 12 24 36 48 60 72 Forecast range (hours)
Abs error of precipitation forecast for grade 1.1-10.5 mm/12 h. July - September 2006. 6,00 EXE NCEP T85 5,00 Reg SLM 4,00 SLMVar 3,00 2,00 1,00 0,00 12 24 36 48 60 72 Forecast range (hours)
EXE Abs. Error of precipitation forecasts NCEP for grade 10.6-19.5 mm/12 h. T85 July-September 2006. 14,00 Reg SLM SLMVar 12,00 10,00 8,00 6,00 4,00 2,00 0,00 12 24 36 48 60 72 Forecast range
Precipitation forecast (accumulated 6hr sum) around Tomsk. Start time 21.07, 00UTC. Valid for 22.07, 07- 13hr local time (left), 13-19 hr (right)
Precipitation forecast (accumulated 6hr sum) around Tomsk. Start time 22.07 00UTC. Valid for 23.07, 07-13hr local time (left), 13-19 local time (right)
Conclusions on precipitation forecasts (1) • In winter, the quality of precipitation forecasts over Western Siberia using SLAV and SLAV-VR models are close to the quality of UKMO and NCEP forecasts up to 48 hr range. Later ranges are somewhat less successful for SLAV models. • The same conclusions are valid for summer, but for 36 hr instead of 48 hrs. Summer conclusions are however preliminary – the period is too short
Conclusions on precipitation forecasts (2) • Quantitatively, the grades more than 10.5 mm/12hr are predicted with larger (negative) mean error as compared with UKMO and NCEP models. For smaller grades , the error is approximately the same. •
Plans for further development of the SL-AV model • Increase of the horizontal resolution first to 0.5, then to 0.25 degrees, 31 then 45 levels in vertical on a new 10 TFlops supercomputer at RHMC. • Implementation of ALARO parameterizations (includes prognostic microphysics, TKE scheme in PBL). + 5 prognostic variables. • Implementation of the reduced grid. • Implementation of the nonhydrostatic dynamical core.
Impact of the horizontal resolution Analysis (color), forecast (isolines). Left - SLAV (70km), right -SLAV-VR (30 km)
Recommend
More recommend
