Centre report of KMA 29th WGNE meeting Melbourne 10-13 MAR. 2014 Hoon Park Numerical Model Management Office KMA.
Download ReportTranscript Centre report of KMA 29th WGNE meeting Melbourne 10-13 MAR. 2014 Hoon Park Numerical Model Management Office KMA.
Centre report of KMA 29th WGNE meeting Melbourne 10-13 MAR. 2014 Hoon Park Numerical Model Management Office KMA Outlines • Current status & update strategy of KMA NWP system • Upgrade NWP system in 2014 and upcoming plans • Research activities in NWP at KMA 2 Status & upgrade strategy KMA’s 3nd Supercomputer (current) Computing System for Operation (20 Cabinets, 379Tflop/s) Ref HaeOn Installation Year Core Number Core Type HaeOn Computing System for R&D and backup (20 Cabinets, 379Tflop/s) HaeDam Total 2010. 12 45,120 45,120 90,240 AMD 2.1 GHz, 12 core Peak performance 379 TF 379 TF 758TF Main Memory 60 TB 60 TB 120TB Capacity of Disk 4 PB Capacity of Tape drive 8 PB OS Suse Linux 11 HaeDam HaeOn, HaeDam was ranked the 110th, 111th fastest supercomputer in the world (Nov. 2013) 4 Ref. ( http://www.top500.org ) Operational NWP Systems GLOBAL • Resolution N512L70 (UM) (~25km / top = 80km) • Target Length 288hrs (00/12UTC) 87hrs (06/18UTC) • Initialization : Hybrid Ensemble 4DVAR E-ASIA •UM 12kmL70 / WRF 10kmL40 • Target Length 87hrs/72hrs (6 hourly) • Initialization : 4DVAR / 3DVAR 5 Global EPS • Resolution N320L70 (UM) (~40km/ top =80km) • Target Length 288hrs • IC : GDAPS • # of Members : 24 • Perturbation : ETKF, RP, SKEB2 LOCAL • Resolution 1.5kmL70 (UM) (744928 / top =39km) • Target Length 36hrs • Initialization : 3DVAR Operational NWP Models (’13.6~) Resolution Target Length Main target N512(25km) L70 12 days 87hours(06,18) Medium-range N320 L70 M24 12 days Medium-range (EPS) UM (E.Asia) 12km L70 87 hours Short-range WRF (E.Asia) 10km L40 87 hours Short-range UM (Korea) 1.5km L70 36 hours Very short-range KLAPS (Korea) 5km 12 hours Very short-range 55km 12days Global 8km 87 hours Northeast Asia 1km 72 hours Coastal ADAM (Dust & Aerosol) 30km 72 hours Asia dust DBAR (Typhoon) 35km 72 hours Track Tide/Storm Surge 9km 87 hours Northeast Asia Model Global E-Asia Local App. & Stat. 6 UM (Global) Wave Watch III Update plans for NWP system Year 2014 2015 computer 3rd 3rd to 4th 2016 2017 4th Global Deterministic 25km 70L Hybrid 4dVar 60km inner loop 17km70L Hybrid 4dVar 40km DA inner loop Global Ensemble 40km 70L 24M 6 hour cycle 25km 70L 24M 6 hour cycle Local (LDAPS) Deterministic 1.5km 70L 3dVar(3hr) 3km inner loop 1.5km 70L 4dVar(1hr) 3km inner loop 1.5km 70L Hybrid 4dVar(1hr) 3km inner loop 1km 70L Hybrid 4dvar(1hr) 2km inner loop Local (LDAPS) Ensemble 3km 70L 12M 3km 70L 12M 3km 70L 12M 1.5km 70L 24M (Semi operation run) (Official operation run) Undetermined Coupling with ocean wave, Asian dust model with global model 7 Update NWP system in 2013 NWP Changes in 2013 Global Data Assimilation and Prediction Syste (N512L70) Version changes UM : vn7.7 → vn7.9 VAR/OPS : vn27.2 →vn29.2 SURF: vn18.2 → vn18.5 Major change 4dVar → Hybrid Ensemble 4dVar Physics package upgrade (PS26 → PS28) Use Climatological Aerosols Data Set Add COMS CSR data 9 NWP Changes in 2013 Regional(East Asia) Data Assimilation and Prediction System(12kmL70) UM : vn7.7 → vn7.9 Physics package: PS27 → PS28 SURF: vn18.5 Ancillary Data Set update (CAP6.6 → CAP7.7) New soil hydraulic properties – wilting and critical points New soil thermal conductivity Local(Korea) Data Assimilation and Prediction System(1.5kmL70) UM : vn7.9 (1.5km L70 ) Physics package : PS27 → PS28 OPS : vn27.2 / VAR : vn27.2 → vn29.2 SURF: vn18.3 → vn18.5 Use Aerosol effect for Visibility with domestic emission data Latent heat nudging using Radar Data 10 Global Hybrid Ensemble 4DVAR 06 UTC 12 UTC 18 UTC 00 UTC Background GDAPS ERLY LATE ERLY LATE ERLY Initial T+0 Obs EPS ERLY 9h ERLY (12d) 288h LATE ERLY LATE BERR ERLY 9h ERLY 288h (12d) • Changes in EPS : 2 times daily (00/12UTC) → 4 times daily (9 hours forecast at 06/18UTC) • Use Hybrid background-error covariance (Climatological covariance : Ensemble covariance = 1.0 : 0.3) to reflect “Error of the day” 11 Impact of Hybrid Ensemble 4DVAR Verification against Observation / Improvement over Non-hybrid D.A. [%] July~August 2012 December 2012 Positive Impact AVG AVG Verification Domain Verification Domain Verification against ECMWF Analysis (Z500) / Improvement over Non-hybrid D.A. [%] July~August 2012 December 2012 Positive Impact NH 12 TR SH NML ASIA Verification Domain EASIA AVG NH TR SH NML ASIA Verification Domain EASIA AVG Radiative Effect of Aerosol (Climatology) Summer Continental Warm Bias 13 Operational Global Model Performance Model : D. A. : GSM T106 1dVar 3DOI (TOVS) 14 GSM T426 GSM T213 3dVar FGAT UM N320 UM N512 4dVar Hybrid 4dVar The effect of New soil hydraulic properties • The corrected soil hydraulic properties data set added to RDAPS(12kmL70). • The new soil properties shows wilting, critical point ↑ → soil moisture ↑ → surface temperature ↓ • Decreased the warm bias in the RDAPS for winter time Old soil properties Comparison of the soil properties OLD New soil properties 15 Day local time Night NEW Visibility with Murk Aerosols (LDAPS) • Aerosol emission of CAPSS1 at NIER2 Resolution : 1km X 1km [ kg/year/km2 ] Type : NOx, SOx, VOC Coverage : South Korea vis 1 CAPSS 2 NIER • : Clean Air Policy Support System : National Institute of Environmental Research INTEX-B (NASA/2006) Aerosols effect Improved underestimation visibility for 1~5km Murk Aerosols Visibility NEW 16 (1 3 p) p 2 m m N r r * 0 air 0 0 m m 0 0 • : visual contrast(=0.05) • air : scattering coefficient of clear air(=1.0E-5) • r* : radius of total water concentration • m : concentration of murk aerosol - L_MURK=F : 10 [g/kg] - L_MURK=T : 0.1~200 [g/kg] Resolution : 0.5o0.5o Type : NOx, SOx, VOC Coverage : East Aisa • ln Relative Humidity OLD Plan for the new HPC introduction HPC Plans - Installation of 1st stage system : ‘14 Q4 - Installation of final stage system : ‘15 Q4 - Expacted Rpeak : ~10PFlops (current HPC : 0.75PFlops) NWP Plans 17 NWP system ‘15~’16 ‘17~’18 Global (Deterministic) ~17km ~12km Global (EPS) ~25km ~17km Local (Deterministic/EPS) 1.5~4km ~1km Future Plans on NWP system Model • Implementation of New Dynamical Core (ENDGame) (’15) • Increasing resolution of the global model(Ocean wave, dust) • Development of Probabilistic (Ensemble) NWP Systems • A-O coupling for NWP (extended medium-range prediction) Data assimilation • Development / Implementation of new D.A. technique - Approach to use Ensemble information in D.A. • Use of additional observation data - CSR, Ground GPS, observation from new satellites • Ocean D.A ( For Seasonal and extended range (30 days)). 18 Seamless Prediction Approach Complexity Earth System Joint Seasonal Prediction System 1day 3days 1km A-O Coupling GAP NWP 10km Atmos.(+Sfc.) Resolution 10days 100km LDAPS RDAPS 3months GDAPS >year Target Length Seasonal Prediction Climate Research Global EPS 19 1month Seamless Prediction Approach Complexity N320(~40km) UM (M24 EPS) + NEMO(0.25deg) 30-day prediction trial Earth System 1day 1km A-O Coupling 10km Atmos.(+Sfc.) Resolution 20 100km 3days 10days 1month 3months >year Target Length NWP Seasonal Prediction Climate Research Extension of target length using coupled NWP system and ensemble approach Spatio-temporally higher resolution prediction Regional & Convective Scale Modelling Incheon UM 1kmL70 for 17th Asian Game (Incheon, 2014) - UM vn7.9 → vn8.2 / 360(E-W)x324(N-S) - Hourly 3DVAR (2km inner loop) cycle / FGAT → 4DVAR? - LBC from LDAPS model - T+12H 21 Incheon Int’l Airport Seoul Local Ensemble & DA research LENS • The integration area covers the Korean peninsula including oceans and parts of adjacent countries such as China and Japan. , 1 86 9 km • About 3km horizontal grid spacing and 70 vertical levels of top 40km altitude are employed. • Simple downscaling of global Ensemble prediction system (N400L70, ~32km) will be adapted for IC and BC. 2,013 km 23 Resolution & Num. of members Preliminary FSS score(Le Duc et al., 2013) result from 3 rainfall cases. [0.1mm] (3km 16) > (3km 12) > (2km 12) ~ (2km 16) > (2km 8) > (3km 8) > (1.5km 16) ~ (1.5km 12) > (1.5km 8) Spatial scale (neighborhood size) [1.0mm] (3km 8) > (3km 12) > (2km 12) ~ (2km 8) ~ (3km 16) > (2km 16) > (1.5km 12) ~ (1.5km 16) > (1.5km 8) Spatial scale (neighborhood size) 3km 16 member shows best 3km 12 member selected to trial accounting computer resources and stable performance 24 Trial schedule of LENS 00 UTC 06 UTC 12 UTC 18 UTC Background GDAPS Hybrid Ensemble and 4dVar EPSG ERLY Initial T+0 Obs. ERLY (12d) Downscaling LENS LATE ERLY LATE LATE ERLY ERLY BG_ERR s ERLY 288h 9h Pert. IC (T+3) Pert. BC ERLY (12d) 288h ERLY 9h Pert. IC (T+3) Pert. BC 03UTC T+48 12~24 members for 2 days forecast 15UTC • Global ensemble prediction system(EPSG) provides perturbed initial and boundary conditions for LENS at T+3 forecast. 25 LATE T+48 Application of Ensemble DA LENS observation sensitivity using LETKF Spin-up 2012. 08. 03. 12 UTC ~ 2012. 08. 04. 12 UTC experiment 2012. 08. 05. 00 UTC ~ 2012. 08. 06. 12 UTC(8 cases) 2012080612 Sonde Surface Aircraft 8192 15415 1390 Sensitivity 4616.63 145.3896 37740.66 Sensitivity/num ber 0.5634 0.009432 2.715155 Forecast error contribution 18428.88 18188.11 257.9684 N. OBS J nObs ( v ) n 1 0 n J v 0 nObs (v ) n 1 0 n J ( v 0 )n :Forecast error contribution :N Obs :Sensitivity 26 Ocean DA and forecasting Purpose and status Objective (application in KMA) Short-range global ocean forecasting Seasonal prediction in KMA (GloSea-5) ocean initial fields Improvement of regional ocean forecasting in KMA lateral condition History Introduction of NEMO-CICE and NEMOVAR from UK Met Office (2012.7) - Pre-operational version of codes • Short-range hindcast simulation - start from 2010/06/10 (currently, running at 2010/08/ ) - using QCed obs. and NWP fluxes of UKMO •Development of pre-processing system (2013.1~12.) - Observations: gathering observations and quality control (NEMOQC) - Fluxes: extraction from KMA NWP and interpolation to model (ORCA025) 28 Hindcast Results 29 10/Jun/2010 ~ 15/Jun/2010 Comparison: SSH 29 Summary & Plan Implementation of NEMO/NEMOVAR at KMA is on going. Next year, works on the post-processing will be conducted we will move a pre-operational this Year. The assessment will also include inter-comparison with other reanalysis data, and comparison to independent data (e.g., surface drifter) The work on diagnosis of the NEMO/NEMOVAR will be started. KMA has a plan to replace a current regional ocean model (ROMS) by regional-NEMO/NEMOVAR system. (northwestern Pacific Ocean with 1/12 deg.) 30 Development at KIAPS 3D global hydrostatic model Cubed Sphere horizontal grid (CAM-SE) Lorenz grid hybrid / Finite difference in vertical(70 layers) ne30np4(~ 1o×1o) Plug-in Selected physics modules & dynamic core in own model framework Develop suitable physics(convection, PBL) around 10 km resolution Jan. 2012 mean zonal wind Jan. 2012 mean precipitation 32 Developing Non-hydrostatic dynamic core • Develop 2-D slice model to test compressible non-hydrostatic equations • Develop 3-D in 2014(IMEX, CG in horizontal, FE in vertical) 33 Data assimilation • Develop basic components for DA – Ensemble DA using LETKF on cubed sphere grid – Minimization & variable conversion using spectral element method for variation method – Developing 4DEnVar following 5 years using developed components – Surface, sonde, AMSU-A, IASI, AIRS data processing developed No DA U V T q Sonde Sonde+AIRS Result from OSSE exp. Using Ensemble DA 34