naval research laboratory marine meteorology division, monterey ca operational application of navdas...
TRANSCRIPT
NAVAL RESEARCH LABORATORYMARINE METEOROLOGY DIVISION, Monterey CA
Operational Application of NAVDAS 3DVAR Analysis for COAMPS
Keith Sashegyi Pat Pauley1, Jason Nachamkin1, Mike Frost1, Chi-Sann Liou1,
Randy Pauley2, Tom Neu2, Dan Geiszler3
1Marine Meteorology Division, Naval Research Laboratory, Monterey, CA2Fleet Numerical Meteorology and Oceanography Command, Monterey, CA
3Science Applications International Corporation, San Diego, CA
Presented at 23rd Conference on Weather Analysis and Forecasting/
19th Conference on Numerical Weather Prediction 4 June 2009, Omaha, Nebraska
OUTLINE
• Introduction• NRL’s NAVDAS analysis
• Update on NAVDAS-AR for NOGAPS• Application to COAMPS®
• Application to COAMPS On-Scene• Tropical Cyclone analysis• Future Plans
NRL Atmospheric Variational Data Assimilation System
• NAVDAS Operational 3DVAR data assimilation system at FNMOC1
– Used for global NOGAPS2 and mesoscale COAMPS®3 models– Running operationally with NOGAPS since October 2003.– Began for the seven different operational COAMPS areas in the period December
2006 (E-PAC) through August 2007 (W_PAC)– Running operationally with COAMPS-OS (CAAPS) on new LINUX cluster at
FNMOC beginning October 15, 2008• NAVDAS Features
– Formulated in observation space – Computational costs are proportional to the number of observations– Uses observations at all levels for profiles (rawinsonde, aircraft ascents)– Vertical mode decomposition of background error for sounding and radiance
observations– Very efficient using soundings and radiances– Unified code for both global and mesoscale NWP systems, with some separate
NOGAPS/COAMPS drivers and routines.
1Fleet Numerical Meteorology and Oceanography Command
2Navy Operational Global Atmospheric Prediction System
3Coupled Ocean Atmosphere Mesoscale Prediction System
Forecast Hour: Analysis 6 hour 12 hour 24 hour
RMS differences between rawinsonde u-component wind observations and NAVDAS / MVOI forecasts for W_ATL 27 km grid for 15-22 Jan 2006
NAVDAS MVOI
NAVDAS/COAMPS®: Improved Fit to Soundings
RMS Differences (m/s)
• NAVDAS includes significant-levels
• MVOI uses only mandatory levels
NAVDAS uses mandatory and significant level winds, temperatures and humidities for rawinsonde soundings. NAVDAS analyses fit observations better over a range of observation types, variables, and vertical levels compared to MVOI.
NAVDAS - AR
• NAVDAS-AR for NOGAPS1
– Weak constraint 4DVAR cast in observation space– Using cycling accelerated representer (AR) method – Flow dependent variation of background error covariance with time– Use all observations in 6 hour time window at time of each observation– New features:
• New preconditioner algorithm reduces computational cost• Adaptive tunnig of observation error varainces
• NAVDAS-AR status– In “Beta” testing with NOGAPS at FNMOC2
– Improved forecast statistics achieved with NOGAPS compared to NAVDAS– Using additional suite of satellite instruments:
• Hyperspectral IR/MW IASI, AIRS; SSMIS microwave; ASCAT scatterometer• Variational Bias correction
• NAVDAS Adjoint for NOGAPS application– Assess observation impact using NAVDAS and NOGAPS adjoint models– Sensitivity of forecast error to observations
1Navy Operational Global Atmospheric Prediction System
2Fleet Numerical Meteorology and Oceanography Command
NRL Global DA group: Liang Xu, Nancy Baker, Jim Goerss, Pat Pauley, Rolf Langland, Bill Campbell, Ben Ruston with Randy Pauley (FNMOC) and Tom Rosmond (SAIC)
NAVDAS-AR 500 mb Height Anomaly Correlation
Comparison of NAVDAS (OPS/L30), NAVDAS-AR with 30 vertical levels (AR/L30) and NAVDAS-AR with
42 vertical levels and model top of 0.04 hPa
NAVDAS-AR Tropical Cyclone Track Verification
Consistently better hurricane forecast tracks; additional testing underway
NAVDAS-AR & NAVDAS Observation Impact
NAVDAS-AR NAVDAS
New satellites: SSMIS, AIRS, IASI
NRL Atmospheric Variational Data Assimilation System
• NAVDAS Application to COAMPS®1
– To calculate innovations (ob-forecast), forecasts from separate COAMPS grid meshes combined to provide single multi-resolution background field
– Utilize NOGAPS2 forecast in halo region of 15 coarse grid points around outer domain
– Consistency of background across grid meshes maintained by feedback of fine grid meshes to coarser grid meshes.
• Full forecast model Two-Way interaction (research mode)• Smoothing differences between grid meshes (ops)
– Variable horizontal correlation length scale (tropical cyclones)
1Coupled Ocean Atmosphere Mesoscale Prediction System
2Navy Operational Global Atmospheric Prediction System
NAVDAS/COAMPS: Multiple grid meshes
Grid 1 Analysis: 81 km Grid 2 Analysis: 27 km
A multi-grid analysis generated by NAVDAS analysis provides the initial conditions for the mesoscale COAMPS weather prediction model. A consistent first-guess forecast field was generated by feedback from the fine grid to the coarse grid and used in NAVDAS with NOGAPS forecast in halo region around coarse grid.
COAMPS/NAVDAS analysis of wind speed (m/s) at 300 mb for E_PAC area
100
90
80
70
60
50
40
30m/s
Feedback of COAMPS Background fields for NAVDAS
COAMPS/NAVDAS analysis of temperature (ºC) and geopotential height (gpm) at 300 mb for E_PAC area
New Grid 1 Analysis: 81 km Grid 2 Analysis: 27 km
Updated NAVDAS/COAMPS grid 1 analysis with feedback of smoothed differences of grid 2 - grid 1 COAMPS forecast fields.
Improved consistency between analyses on the separate grids, when using single multi-grid NAVDAS analysis.
NAVDAS for COAMPS-On Scene
• COAMPS-OS provides user friendly web-based GUI control for setting up and running COAMPS modeling system
– Grid meshes set up and system run on demand, not with fixed job schedule– Utilize latest available NOGAPS forecasts and observations
• Integration of NAVDAS into COAMPS-OS system– Ported and tested NAVDAS with COAMPS-OS (CAAPS) on new Linux cluster
at FNMOC – More adaptable & robust for variety of grid configurations and run environments– More efficient use of local disks on LINUX cluster nodes– More portable to various computer platforms (LINUX, SGI, IBM)– Results reproducible for different number processors using double precision
• NAVDAS now running operationally with COAMPS-OS (CAAPS) on LINUX cluster at FNMOC since October 15, 2008
– 3 hourly NOGAPS forecast boundary conditions– 30 pressure levels for analysis and NOGAPS BC’s
Integration of NAVDAS/COAMPS with COAMPS-OS
COAMPS-OS/NAVDAS comparison with COAMPS-OS/MVOI
MVOI Analysis: 54 km NAVDAS Test Analysis: 54 km
NAVDAS/COAMPS software integrated into COAMPS On-Scene system. More consistent features analyzed with NAVDAS.
COAMPS-OS controlled by web-based interface, with easy setup of different grid configurations, web display of COAMPS fields. Tested various grid configurations (across Dateline, Greenwich Meridian, small areas) on NRL Linux computers.
Integration of NAVDAS with COAMPS-OS at FNMOC
NAVDAS Analysis and COAMPS-OS forecast on operational 27 km EPAC domain
Testing and validation of NAVDAS/COAMPS-OS cycling on FNMOC LINUX cluster. NAVDAS has been running operationally with COAMPS-OS at FNMOC since October 15, 2008.
00Z Nov 4 2008
Analysis 250mb Wind Speed 12 hr forecast slp & total precipitation
12Z Nov 3 2008
NAVDAS Data Assimilation for Tropical Cyclones
• NAVDAS provides the capability for reducing the background correlation length scale for the analysis of smaller scale systems such as tropical cyclones.
• In the vicinity of tropical cyclones, reducing both the correlation length scale and the geostrophic coupling of the wind and height, with the prior relocation of the forecast tropical cyclone, results in improved analyses for providing the initial conditions for forecasting of tropical cyclones.
• Experience with running NAVDAS for many TC cases during T-PARC/TCS08, has resulted in tuning of NAVDAS to handle both weak and strong tropical cyclones.
• TC version of COAMPS and NAVDAS to be run in “beta” test mode at FNMOC this summer.
• Ongoing testing further modifications of NAVDAS covariances in the region of a tropical cyclone.
Tropical Cyclone Analysis with NAVDAS/COAMPS
Sea level pressure Wind Speed /Direction at 850 mb
Tropical Cyclone Isabel - NAVDAS Analysis
Analysis central pressure: 956 mb
Observed: 942 mb
Analysis max wind: 67 m/s
Observed: 60 m/s
NAVDAS has been adapted for use with COAMPS for Tropical Cyclone analysis. The predicted tropical cyclone is relocated to correct the position. A reduced correlation length scale and reduced geostrophic coupling are then used. A more realistic tropical cyclone is provided for the initial conditions for COAMPS.
Tropical Cyclone Analysis with NAVDAS/COAMPS
Analysis of 1000mb wind speed (m/s) and height for weak TC case
NAVDAS Analysis: L=80 km
NAVDAS/COAMPS analyzes tropical cyclone bogus observations of wind, temperature and 1000mb height with a reduced correlation length scale and reduced geostrophic coupling in TC. For weak systems with ill-defined circulations, analysis is very sensitive to the correlation length scale. To handle both strong and weak systems, a length scale of 185 km with geostrophic coupling of ¼ value outside TC is now used.
NAVDAS Analysis: L=200 km
1002 mb 1004 mb
August 18 2008
Tropical Cyclone Analysis with NAVDAS/COAMPS
Analysis of 850 mb wind speed (m/s) and sea-level pressure (mb)
Strong TC: Sinlaku Sept 11, 2008
NAVDAS/COAMPS analyzes of tropical cyclone bogus observations of wind, temperature and 1000mb height with a reduced correlation length scale and reduced geostrophic coupling in TC. Better fit to observed wind speed than observed sea-level pressure.
1004 mb
JMA ~ slp/JTWC max wind: 935 mb/ 60 m/s
961 mb
JMA ~ slp/JTWC max wind: 996 mb /25 m/s
Weak TC: Nuri, Aug 18, 2008
Future Plans• Testing of TC modifications to NAVDAS and COAMPS
– “Beta” test at FNMOC this summer• Update satellite derived feature-track winds processing fpr COAMPS
– Implement MODIS & WINDSat winds– GOES rapid-scan imagery
• Adapt NAVDAS to generate hourly analyses using higher frequency observations– Incorporate NAVDAS innovation statistics for data monitoring– Improve space and time interpolation of COAMPS forecasts to observation
time and location for handling of more frequent updates• Complete pre-processing testing for radiance assimilation with COAMPS
– Adapt global QC and bias correction for COAMPS– Develop bias correction strategies for mesoscale