Overview of Rutgers Ocean Modeling Group activities with 4DVar data assimilation in the Mid-Atlantic Bight John Wilkin NOS Silver Spring Feb 28-29, 2012 Apr 24-25, 2013
LaTTE domain and observation locations. Bathymetry of the New York Bight is in grayscale and dashed contours. Yellow star is location of Ambrose Tower Green squares are CODAR HF Radar sites LaTTE 2006 Data Assimilative reanalysis (60 days) Zhang, W., J. Wilkin and H. Arango (2010a), Towards an integrated observation and modeling system in the New York Bight using variational methods, Part I: 4DVAR Data Assimilation, Ocean Modelling, 35, , doi: /j.ocemod Zhang, W., J. Wilkin and J. Levin (2010b), Towards an …, Part II: Representer-based observing system design, Ocean Modelling, 35, , /j.ocemod
We overlap 3-day analysis cycles, performing a new analysis and new forecast every day
Comparison of observed and modeled sea surface temperature and current at 0700 UTC 20 April LaTTE 2006 reanalysis (60 days) Zhang, W., J. Wilkin and H. Arango (2010a), Towards an integrated observation and modeling system in the New York Bight using variational methods, Part I: 4DVAR Data Assimilation, Ocean Modelling, 35, , doi: /j.ocemod Zhang, W., J. Wilkin and J. Levin (2010b), Towards an …, Part II: Representer-based observing system design, Ocean Modelling, 35, , /j.ocemod
2-D histograms comparing observed and modeled temperature, salinity, and u- component of velocity model before (control simulation) and after (analysis) data assimilation. Color indicates the log 10 of the number of observations. LaTTE 2006 reanalysis (60 days) Zhang, W., J. Wilkin and H. Arango (2010a), Towards an integrated observation and modeling system in the New York Bight using variational methods, Part I: 4DVAR Data Assimilation, Ocean Modelling, 35, , doi: /j.ocemod Zhang, W., J. Wilkin and J. Levin (2010b), Towards an …, Part II: Representer-based observing system design, Ocean Modelling, 35, , /j.ocemod
Surface velocity forecast skill improves with assimilation of CODAR This analysis for ROMS LaTTE domain (NY Bight). 2-day forecast skill significantly improved for cross-correlation (submesoscale pattern variability) MSE = mean squared error CC = cross-correlation; Sm and So are std. dev. of model and obs analysis forecast window analysis forecast window
4DVar Assimilation (physics) in ROMS ESPreSSO* operational system for OOI CI OSSE (ongoing) – 72-hour forecast (NAM-WRF meteorology) – tides, rivers, OBC HyCOM NCODA – assimilates: altimeter along-track SLA satellite IR SST CODAR surface currents climatology glider T,S GTS: XBT/CTD, Argo, NDBC *Experimental System for Predicting Shelf and Slope Optics ESPreSSO 7
*Experimental System for Predicting Shelf and Slope Optics ESPreSSO Work flow for operational ESPreSSO/MARCOOS 4DVar ROMS 4DVAR Analysis and Forecast Analysis interval is 00:00 – 24:00 UTC Input pre-processing starts 01:00 EST Input preprocessing completes approximately 05:00 EST 4DVAR analysis completes approx 08:00 EST 24-hour analysis is followed by 72-hour forecast using NCEP NAM 0Z cycle available from NOMADS OPeNDAP at 02:30 UT (10:30 pm EST) Forecast complete and transferred to OPeNDAP by 09:00 EST Effective forecast is ~ 60 hours
Work flow for operational ESPreSSO/MARCOOS 4DVar Data used: 72-hour forecast (NAM-WRF meteorology 0Z cycle at 2 am EST) RU CODAR is hourly - but with 4-hour latency delay RU glider T,S when available (seldom) (~ 1 hour delay) USGS daily average flow available 11:00 EST – persist in forecast AVHRR IR passes 6-8 per day (~ 2 hour delay) HyCOM NCODA 7-day forecast updated daily Jason-2 along-track SLA via RADS (4 to 16 hour delay for OGDR) – Also ENVISAT and Jason-1 NRT data (OGDR and IGDR) SOOP XBT/CTD, Argo floats, NDBC buoys via GTS from AOML Regional high-resolution T,S climatology (MOCHA*) *Mid-Atlantic Ocean Climatology Hydrographic Analysis
Work flow for operational ESPreSSO/MARCOOS 4DVar Input pre-processing RU CODAR de-tided (harmonic analysis) and binned to 5km – variance within bin & OI combiner expected u_err (GDOP) used for QC >> ROMS tide added to de-tided CODAR – reduces tide phase error contribution to cost function RU glider T,S averaged to ~5 km horiz. and 5 m vertical bins – need thermal lag salinity correction to statically unstable profiles SOOP XBT and Argo – not used at present AVHRR IR individual passes 6-8 per day – U. Del cloud mask; bin to 5 km resolution – REMSS daily SST OI combination of AVHRR, GOES, AMSR-E Jason-2 along-track 5 km bins (with coastal corrections) from RADS – MDT from 4DVAR on “mean model” (climatology 3D T,S, u CODAR, τ wind ) >> add ROMS tide solution to SSH USGS daily river flow is scaled to account for un-gauged watershed
ESPreSSO operational system 11
ESPreSSO operational system
14 Skill of DA analysis at hind-casting mesoscale SST
15 Skill of DA analysis at hind-casting along-track SSHA Jason-2 Envisat
16 Comparison to withheld T and S observations from CTD, gliders and XBT 16
17 Skill at hind-casting vertical structure of salinity. These observations were not assimilated
18 Skill at hind-casting vertical structure of temperature. These observations were not assimilated
ROMS LaTTE and ESPreSSO 4DVAR use all available data from a modern coastal ocean observing system – satellites, HF-radar, moorings, AUV (glider, Argo …), XBT/CTD; IR SST individual passes work best; time variability resolved – more and diverse data is better – climatology assimilation: removes OBC bias; improves representation of dynamic modes and adjoint-based increments Useful skill for operational applications – 5-7 days for temperature and salinity – 1-2 days for velocity – improved short-term ecosystem prediction – observing system operation … glider path planning Variational methods for observing system design – adjoint sensitivity and representer-based observing system design (see W. Zhang et al. papers in Ocean Modelling, 2010); observation impact analysis (see A. Moore et al. papers in Prog. Oceanog. 2011) Summary
Switch to ROMS 4DVAR formulation based on weak constraint/dual space formulation Solution is in “observation” space – typically much much smaller than model space Uses same forward, adjoint and tangent linear models, and observation input format Comparable convergence and speed Two variants: – #define W4DVAR - Indirect Representer algorithm (Egbert et al. 1994) – #define W4DPSAS - Physical Space Statistical Analysis System (Da Silva et al. 1995) Allows for adjustment to time-varying forcing and boundary conditions, explicit acknowledge of model errors, and posterior analysis of e.g. observation impact (and more) DOPPIO model domain configuration with local shelf and estuarine nests Future Doppio …
Proposed Doppio domain with Rio09 Mean Dynamic Topography Grid cell resolution 7 km
Proposed IODA domain (version 3) (only every 3 rd grid cell) to nest within Doppio Grid cell resolution 2.33 km Lm x Mm 96 x 144 the resolution challenge …
RU Endurance Line glider transect May 18-24, 2006 Small scales – important to large scale dynamics (and ecosystem)?