Скачать презентацию HYCOM NCODA Variational Ocean Data Assimilation System James Cummings Скачать презентацию HYCOM NCODA Variational Ocean Data Assimilation System James Cummings

00cbd1656df64341569edb9c790ab399.ppt

  • Количество слайдов: 20

HYCOM/NCODA Variational Ocean Data Assimilation System James Cummings Naval Research Laboratory, Monterey, CA GODAE HYCOM/NCODA Variational Ocean Data Assimilation System James Cummings Naval Research Laboratory, Monterey, CA GODAE Ocean View III Meeting -18 November 2011 European Space Agency Headquarters Paris, France 14

NCODA: Variational Analysis Flexible and Unified System: • global or regional applications (HYCOM, NCOM, NCODA: Variational Analysis Flexible and Unified System: • global or regional applications (HYCOM, NCOM, WW 3) • 2 D mode: SST, Sea Ice, SSH, SWH, surface velocity • 3 D mode: fully multivariate analysis (T, S, U, V) • multi-scale analyses: nested, successively higher resolution grids • cycles with forecast model or runs stand-alone Designed as Complete End-to-End Analysis System: • data quality control (QC) • variational analysis (3 DVAR) • performance diagnostics (analysis residuals, Jmin, adjoint data impacts, ensemble transform)

HYCOM/NCODA Data Flow Raw Obs SST: NOAA (GAC, LAC), METOP (GAC, LAC), GOES, MSG, HYCOM/NCODA Data Flow Raw Obs SST: NOAA (GAC, LAC), METOP (GAC, LAC), GOES, MSG, MTSAT-2, AATSR, AMSR-E, Ship/Buoy in situ Profile Temp/Salt: XBT, CTD, Argo Floats, Fixed/Drifting Buoy, Ocean Gliders NCODA: Navy Coupled Ocean Data Assimilation Automated QC w/condition flags Ocean Data QC 3 DVAR – simultaneous analysis of 5 ocean variables: temperature, salinity, geopotential, u, v velocity components Innovations Altimeter SSH: Jason-1&2, ENVISAT Sea Ice: SSM/I, SSMIS, AMSR-E 3 DVAR Increments Velocity: HF Radar, ADCP, Argo Trajectories, Surface Drifters, Gliders Adaptive Sampling Data Impacts Sensors Forecast Fields Prediction Errors NCODA: QC + 3 DVAR HYCOM First Guess

NCODA Analysis System Components • 3 DVAR • Analysis Error • Ensemble Transform • NCODA Analysis System Components • 3 DVAR • Analysis Error • Ensemble Transform • Assimilation Adjoint (KT)

NCODA: Data Impacts Analysis – Forecast System Observation (y) Background (xb) NCODA 3 DVAR NCODA: Data Impacts Analysis – Forecast System Observation (y) Background (xb) NCODA 3 DVAR Analysis (xa) HYCOM / NCOM Forecast (xf) Adjoint System Ob Error Sensitivity ( J/ e) Observation Sensitivity ( J/ y) Adjoint of NCODA 3 DVAR Observation Impact ( J/ y) How to adjust the specified errors to improve the forecast ? Analysis Sensitivity ( J/ xa) HYCOM / NCOM Adjoint Gradient of Cost Function J: ( J/ xf) What is the impact of observations on the forecast accuracy ?

NCODA: SST Data Impacts Analysis – Forecast System Observation (y) Background (xb) NCODA 2 NCODA: SST Data Impacts Analysis – Forecast System Observation (y) Background (xb) NCODA 2 DVAR Analysis (xa) Navy NWP (NOGAPS) Forecast (xf) Adjoint System Observation Sensitivity ( J/ y) Adjoint of NCODA 2 DVAR Analysis Sensitivity ( J/ xa) Navy NWP Adjoint What is the sensitivity of the low level wind stress to the different SST data sources ? Gradient of Cost Function J: ( J/ xf)

NCODA: SST Data Sources • GOES 11, 13 (NAVO) • MSG (GHRSST GDAC) • NCODA: SST Data Sources • GOES 11, 13 (NAVO) • MSG (GHRSST GDAC) • METOP GAC/LAC (NAVO) • NOAA 18, 19 GAC/LAC (NAVO) • Drifting/Fixed Buoys • Ship intake, hull contact, bucket temps Coming Soon: • MTSAT-2, NPP VIIRS, Wind. SAT

NCODA: Adaptive Data Thinning • high density surface data averaged within spatially varying bins NCODA: Adaptive Data Thinning • high density surface data averaged within spatially varying bins – applied to SST, SSH, SWH, HF Radar, sea ice data • bins defined by grid mesh and background covariance structure – more (less) thinning where length scales are long (short) • takes into account observation error and SST water mass of origin Thinned SST Global NWP 37 km grid Length Scales 10 km 200 km 10 km input # obs: 28, 943, 383 output # obs: 152, 768

NCODA: Direct Assimilation Satellite SST Radiances Assume changes in TOA radiances are due to: NCODA: Direct Assimilation Satellite SST Radiances Assume changes in TOA radiances are due to: (1) atmospheric water vapor content (2) atmospheric temperature (3) sea surface temperature Channel 3: 3. 5 m Channel 4: 11 m Channel 5: 12 m CRTM provides sensitivity of radiances with respect to SST, water vapor, and atm temperature for SST channels

NCODA: Assimilation Satellite SST Radiances Given TOA BT innovations and RTM sensitivities, solve: Returns: NCODA: Assimilation Satellite SST Radiances Given TOA BT innovations and RTM sensitivities, solve: Returns: (1) SST increment - Tsst (2) atmospheric temperature increment - Tatm (3) atmospheric moisture increment - Qatm • incorporates impact of real atmosphere above the SST field • removes atmospheric signals in the data • knowledge of esst, eq error statistics critical

NCODA: Assimilation Satellite SST Radiances • δTSST corrections for NOAA-19 and METOP-A; valid 8 NCODA: Assimilation Satellite SST Radiances • δTSST corrections for NOAA-19 and METOP-A; valid 8 June 2011 NOAA-19 • first guess SST from NAVO empirical buoy match up regressions • atmos profiles from Navy NWP • large SST corrections associated with high water vapor regions • corrections differ between NOAA-19 and METOP-A for same NWP fields METOP-A

Difference between 2 DVAR analysis of atmospheric corrected and uncorrected NAVO SST - 16 Difference between 2 DVAR analysis of atmospheric corrected and uncorrected NAVO SST - 16 Aug 2011: METOP-A, NOAA-18, 19 • NAVO SST data biased cold • large bias in mid-latitudes during NHEM summer • Atmosphere corrected SST being tested in Navy NWP 4 DVAR • More accurate ocean surface allows use of sounder channels in 4 DVAR that peak in boundary layer • Better characterization of boundary layer will improve ocean forcing

NCODA: Global HYCOM • basin scale assimilation in Mercator part of grid (Atlantic, Indian, NCODA: Global HYCOM • basin scale assimilation in Mercator part of grid (Atlantic, Indian, Pacific) • Arctic cap basin for irregular bi-pole part of grid (not shown) Observation Locations: 4 September 2008 369, 593 obs 263, 427 obs 625, 359 obs

NCODA: Global HYCOM Assimilation Timings on Cray XTE Domain Grid Size Number Procs Number NCODA: Global HYCOM Assimilation Timings on Cray XTE Domain Grid Size Number Procs Number Obs Solver (min) Post (min) Total (min) Atlantic 1751 x 1841 x 42 104 269, 593 1. 3 2. 6 4. 9 Indian 1313 x 1569 x 42 88 263, 427 1. 4 2. 9 4. 5 Pacific 2525 x 1841 x 42 392 625, 359 2. 1 1. 5 4. 0 Arctic Cap 4425 x 1848 x 42 40 181, 230 0. 8 2. 4 3. 2 >750 million grid nodes, ~1. 2 million observations, ~5 min run time

NCODA: HYCOM Verification Temperature Atlantic Indian Pacific model errors adjust to data after ~10 NCODA: HYCOM Verification Temperature Atlantic Indian Pacific model errors adjust to data after ~10 cycles, remain constant over time

NCODA: HYCOM Verification Salinity Atlantic Indian Pacific little model error adjustment to data, Atlantic NCODA: HYCOM Verification Salinity Atlantic Indian Pacific little model error adjustment to data, Atlantic salinity errors worse

NCODA: HYCOM Verification Layer Pressure Atlantic Indian Pacific model errors adjust to data in NCODA: HYCOM Verification Layer Pressure Atlantic Indian Pacific model errors adjust to data in about month slow improvement over time in Atlantic and Indian basin RMS errors

NCODA: First Guess at Appropriate Time Why FGAT? Eliminates component of analysis error that NCODA: First Guess at Appropriate Time Why FGAT? Eliminates component of analysis error that occurs when comparing observations and forecasts not valid at same time -12 0 12 Data Window (+/- 12 hours) 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 Forecast Time Period Innovations 1 hour forecast interval for SST: preserves diurnal cycle

NCODA: First Guess at Appropriate Time 0 -120 12 Data “Receipt Time” Window (-120 NCODA: First Guess at Appropriate Time 0 -120 12 Data “Receipt Time” Window (-120 to + 12 hours) 24 Hour Forecast 5 days ago 24 Hour Forecast 4 days ago 24 Hour Forecast 3 days ago 24 Hour Forecast 2 days ago 24 Hour Forecast 1 day ago Innovations 24 hour forecast interval for profiles assimilating data “received” since last analysis using forecasts valid 5 days into the past

Questions ? Questions ?