Progress and Prospects of U . S . Data Assimilation in Ocean Research

Abstract : This report summarizes goals, activities, and recommendations of a workshop on data assimilation held in Williamsburg, Virginia on September 9-11, 2003, and sponsored by the U.S. Office of Naval Research (ONR) and National Science Foundation (NSF). The overall goal of the workshop was to synthesize research directions for ocean data assimilation (DA) and outline efforts required during the next 10 years and beyond to evolve DA into an integral and sustained component of global, regional, and coastal ocean science and observing and prediction systems. The workshop built on the success of recent and existing DA activities such as those sponsored by the National Oceanographic Partnership Program (NOPP) and NSF-Information Technology Research (NSF-ITR). DA is a quantitative approach to optimally combine models and observations. The combination is usually consistent with model and data uncertainties, which need to be represented. Ocean DA can extract maximum knowledge from the sparse and expensive measurements of the highly variable ocean dynamics. The ultimate goal is to better understand and predict these dynamics on multiple spatial and temporal scales, including interactions with other components of the climate system.

DA is a quantitative approach to optimally combine models and observations.The combination is usually consistent with model and data uncertainties, which need to be represented.Ocean DA can extract maximum knowledge from the sparse and expensive measurements of the highly variable ocean dynamics.The ultimate goal is to better understand and predict these dynamics on multiple spatial and temporal scales, including interactions with other components of the climate system.There are many applications that involve DA or build on its results, including: coastal, regional, seasonal, and inter-annual ocean and climate dynamics; carbon and biogeochemical cycles; ecosystem dynamics; ocean engineering; observing-system design; coastal management; fi sheries; pollution control; naval operations; and defense and security.These applications have different requirements that lead to variations in the DA schemes utilized.For literature on DA, we refer to Ghil and Malanotte-Rizzoli (1991), the National Research Council (1991), Bennett (1992), Malanotte-Rizzoli (1996), Wunsch (1996), Robinson et al. (1998), Robinson and Lermusiaux (2002), and Kalnay (2003).We also refer to the U.S. Global Ocean Data Assimilation Experiment (GODAE) workshop on Global Ocean Data Assimilation: Prospects and Strategies (Rienecker et al., 2001); U.S. National Oceanic and Atmospheric Administration-Offi ce of Global Programs (NOAA-OGP) workshop on Coupled Data Assimilation (Rienecker, 2003); and, NOAA-NASA-NSF workshop on Ongoing Analysis of the Climate System (Arkin et al., 2003).

Climate-System Time Scales
Due to its high capacity to store and redistribute heat, the ocean plays a cen-  (1) carry out coupled acoustical-physical-biogeochemical-hydrological DA; (2) derive fundamental or simplifi ed

PRESENT STATUS OF OCEAN DATA ASSIMILATION
The workshop presentations included presentations on DA systems and new projects.These presentations are summarized below.

New Research Projects
The   Ensembles can be generated from initial

RECOMMENDATIONS AND SYNTHESIS OF PLENARY DISCUSSIONS
Plenary discussions and working groups identifi ed major issues, emerging challenges, and future directions for ocean DA research and operations.These discussions and reports were synthesized to lead the following set of major and immediate recommendations.

I. Ocean and Climate Sciences and Operations
Foster the scientifi c use of DA, with specifi c emphases on multiple scales, coupled ocean-atmosphere and land-surface processes, and interdisciplinary ocean science.
Ocean-state and parameter estimation via DA are carried out in several pilot settings, employing a variety of approaches and using the available data as constraints.Results will continue to be improved over the years to come, but many of such synthesis and reanalysis products are of suffi cient quality to be used for scientifi c studies.

II. Hypotheses Testing and Uncertainty
Foster hypothesis-testing DA research to evaluate and improve models, estimate data and model uncertainties, and assess results. •

IV. Observations and Observing Systems
Bolster rigorous DA research and applications to optimize the design of observing systems, control the quality of measurements and adjust observational plans through adaptive sampling.
• Observing system optimization.In T H I S R E P O R T summarizes goals, activities, and recommendations of a workshop on data assimilation held in Williamsburg, Virginia on September 9 -11, 2003, and sponsored by the U.S. Offi ce of Naval Research (ONR) and National Science Foundation (NSF).The overall goal of the workshop was to synthesize research directions for ocean data assimilation (DA) and outline efforts required during the next 10 years and beyond to evolve DA into an integral and sustained component of global, regional, and coastal ocean science and observing and prediction systems.The workshop built on the success of recent and existing DA activities such as those sponsored by the National Oceanographic Partnership Program (NOPP) and NSF-Information Technology Research (NSF-ITR).
To help the planning of sustained DA research, NSF and ONR requested a synthesis of current efforts and major research directions, including strategies for transitioning research results into M E E T I N G R E P O R T Progress and Prospects of U.S. Data Assimilation in Ocean Research sustained activities that are useful for the whole community.Discussions were therefore centered on the needs of the community for all temporal and spatial scales; current status of ocean DA; major science issues and future challenges; transitions from the present status to sustained activities; and required funding and infrastructure issues.Both research and operational needs were addressed.The specifi c objectives of the workshop were to: 1. Assess the state of the art in DA and discuss the research required to realize its full potential.2. Summarize the status of major national ocean DA activities and enuthe needs for education and training and for retaining scientists involved in DA.More details are provided at http://www.atmos.umd.edu/%7Ecarton/dameeting/abstracts.htm.
tral role in climate variability on time scales of tens to hundreds of years.To understand past and predict future climate changes, reasonable ocean estimates are necessary.Many key climate science questions (e.g., carbon cycling) involve interactions between the physical state of the ocean and its biogeochemistry.On global climatic scales, coupled oceanic and atmospheric processes are also important; it is crucial to refi ne and test air-sea fl uxes to improve coupled models.Realistic climate studies require a synthesis of all available and diverse data sets into a description of the fourdimensional evolution of the ocean consistent with a numerical dynamical model and its uncertainties.With such rigorous DA, momentum, heat, and salt are conserved.Major research issues in DA for climate purposes include: (1) inherent biases in the model equations, numerical representations, multiple-scale data and observation operators; (2) errors of representativeness, or the errors in the data due to the presence of variability on scales distinct from climatic scales; (3) improvement of predictive skill by DA of mesoscale features into climate models; and (4) inference of sub-surface ocean fi elds from the fourdimensional DA of surface (satellite) information only.Seasonal to Decadal Time Scales At these scales, four processes have received special attention for fundamental and societal reasons.First, climate in many parts of the world is linked to the El Niño-Southern Oscillation (ENSO) in the tropical Pacifi c and Indian Oceans.Second, eastern North America and western Europe are infl uenced by a lowfrequency oscillation in the latitude and intensity of Atlantic Ocean storms.Long multi-year fl uctuations of this North Atlantic Oscillation (NAO) are linked to interactions between the atmosphere and the thermohaline circulation of the ocean.Third, the tropical Atlantic Ocean is subject to long multi-year variability (Tropical Atlantic Variability [TAV]) that depends on heat exchanges among the near-equatorial, tropical, and extratropical regions, and affects the eastern tropical Americas and western Africa.Fourth, the Pacifi c Decadal Oscillation (PDO) infl uences the climate of the North Pacifi c Ocean.It involves oceanatmosphere interactions and changes in properties of major wind systems and of ocean variables such as sea surface temperature (SST).
model the sources of uncertainties in the multivariate data and models; and (4) estimate of the most useful data via adaptive sampling and observationsystem simulation experiments.
goal of the HYbrid Cordinate Ocean Model (HYCOM, which is part of GODAE) multi-institution partnership is to develop and apply eddy-resolving, real-time global and basin-scale prediction systems using HYCOM and to transition these systems to the U.S. Navy and NOAA (Chassignet et al., this issue).As part of HYCOM, a Hybrid Ocean Modeling Environment (HOME) will be developed and several ocean DA systems will be implemented and utilized.The fi rst operational assimilation system will be a multivariate optimum interpolation (MVOI), with subsequent evaluations of the Reduced Order Information Filter (ROIF) and Singular Evolutive Extended Filter (SEEK).Atlantic and global confi gurations of the system will be used for varied applications.Products will be evaluated by comparison with independent data.The coastal ocean community is expected to use and evaluate products.Research directions will include correction of mean biases (e.g., sea surface height [SSH]), parameterizations (e.g., sea ice), grid resolutions, DA of new data types (e.g., SST, profi les, drifters), and product distribution.NOPP established the Estimating the Circulation and Climate of the Ocean (ECCO) consortium (Figure 1) to bring ocean-state estimation to a quasi-operational tool for studying large-scale ocean dynamics, designing observational strate-gies, and examining the ocean's role in climate variability.Important activities include re-analyses and initializations for forecasting.Model-data misfi ts are reduced based on adjoint assimilation and Kalman fi ltering and smoothing.The initial conditions and forcing are the control variables.Re-analysis fi elds for 11 years at one degree are available.A 50-year analysis that includes mixing as a control parameter is underway.The reduction of model-data misfi ts is estimated for each data type and the variability and mean of models and data are compared.Errors are small in the open ocean but are larger in western boundary regions.Research directions include comparisons to other products, time-dependent surface fl uxes and transports, higher-resolutions including eddy-resolving DA, coupled and biogeochemical modeling, data and model error statistics, extended control space, parameterizations, observing system design, and sustained almost realtime global estimations.The Fleet Numerical Meteorology and Oceanography Center (FNMOC) is one of the principal weather and ocean prediction centers within the U.S. Department of Defense (DOD).It runs a suite of global (NOGAPS, EFS, WW3, OTIS, TOPS) and regional (COAMPS, WW3, GFDN, PIPS) atmospheric and oceanographic modeling systems.Upcoming model improvements include increased resolutions, nesting, variational assimilation, boundary conditions, and new parameterizations.New parallel computing hardware (at the High-Performance

Figure 1 .Figure 2 .
Figure 1.Mean fl ow fi eld at 27 m and 1975 m depth from the 1 degree Estimating the Circulation and Climate of the Ocean (ECCO) synthesis calculation, together with the mean sea surface height (SSH) and the temperature fi eld at 1975 m.Th e full time-dependent circulation is simulated by the ECCO eff ort since the beginning of 1992 in a way that is consistent with the surface forcing and the model dynamics.Although results are smooth due to the model resolution, the models transport estimates are optimally constrained by all available in situ and satellite data.

Figure 3 .
Figure 3.Comparison of 100 m temperature from three reanalyses in the western tropical Atlantic for April, 1991.Temperature variability is introduced by northwestward moving warm core eddies containing water originating in the southern hemisphere.Diff erences in the representation of the eddies in three reanalyses, one from GFDL, one from NCEP, and one from University of Maryland (SODA, Carton and Giese, 2006) illustrate the challenge of producing accurate re-analyses of the mesoscale ocean prior to the launch of the Topex/ Poseiden altimeters in 1992.
schemes, and new adaptive modeling and sampling.The legacy systems are the Harvard Ocean Prediction System (HOPS), Error Subspace Statistical Estimation (ESSE) and Massachusetts Institute of Technology (MIT) acoustical models.They are encapsulated at the binary level using XML and I/O workfl ows.ESSE is used for physical-acoustical and physical-biological DA.Adaptive

Figure
Figure 4. Organization chart of the Inverse Ocean Modeling (IOM) System.Courtesy of J. Muccino (Arizona State University) and A. Moore (University of California, Santa Cruz).

Figure 5 .
Figure 5. Schematic of the architecture of the Littoral Ocean Observing and System (LOOPS/Poseidon)(Patrikalakis et  al, 2006).Remote compute and data resources, as well as adaptive observation systems, are connected to users through the use of grid computing Simulations and forecasts can be driven from a web browser and a portal gateway to the grid.data for the software are written and utilized for web access.Ocean DA is carried out using the Harvard Ocean Prediction System (HOPS) and Error Subspace Statistical Estimation (ESSE).Th e repository for the metadata of ocean data and the manager of storage resources for keeping track of fi le locations are colored in light blue to indicate that they have not been developed.

Figure 6 .
Figure6.Th e Environmental Data Assimilation Project is illustrated by results of an estimation of the tropical Atlantic circulation from altimetry data using the ensemble Kalman fi lter(Zang and Malanotte-Rizzoli, submitted).Shown are the temporal evolution of sea surface height anomaly (ssha) anomaly from the observations (red line), the control run without DA (black line), and the mean of the analysis (green line) at two selected locations.Th e green lines are the ensemble analyses.
Multiple scales.More efforts are needed on DA for the multiple and coupled oceanic scales, from the interactions of large-scales, mesoscales, and coastal scales, to turbulence effects in estuaries.Some of the generic research involves: (1) understanding multi-scale interactions, (2) inferring and modeling the effects of smaller scales in larger-scale simulations, (3) comparing the benefi ts of enhanced model resolution and model physics to those of enhanced estimation methods, (4) developing effi cient DA schemes that explicitly address multiple scales and strong nonlinearities, and (5) developing multi-method DA approaches as well as DA schemes that learn from data.Examples of such research are quantifi ying tidal-mesoscale interactions, fi nding adequate parameterizations of the eddy fi eld in global and climate estimation, and nested modeling with scale-depend dynamics.• Oceanic-atmospheric-land processes.Coupled DA will be essential for accurate simulations of such processes (e.g., CO 2 sequestration in the ocean).Generic research challenges include identifying the sources of errors in the coupled models and the data, and carrying-out effi cient DA with compre-DA are a venue of the future.Some of the generic research involves: (1) using DA for fundamental research and process discovery in interdisciplinary ocean and climate science, (2) using DA to estimate the parameters and also the mathematical structures of biogeochemical models, (3) developing multi-method and adaptive DA for interdisciplinary studies, and (4) establishing the merits of nested or variable grids in ecosystem (e.g., for biological thin-layer dynamics) and acoustical modeling (e.g., for monitoring).
the past observing systems were set empirically.Through DA, one can now evaluate individual data sets and their role in constraining the estimate and helping our understanding of the ocean.Such optimization of the observing system using rigorous DA must be sustained.This allows revisiting the observing strategy and continuously adapting the network to the evolving dynamics as well as to new science questions.•Ocean data quality control.Effective quality control of ocean data is essential.The use of model predictions as proxies in the quality control process is promising.Comparing models to data in real-time as the data are collected can also help resolve issues of model skill, bias and lack of variability.•Adaptive sampling.The path, locations and other properties of observing platforms and sensors can be optimized and adapted in real-time, so as to respond to the ocean variability and its uncertainties.V. Technology, Systems, and TransfersLink DA efforts to cyberinfrastructure initiatives and increase the support for sustained and effi cient cooperative transitions of methodologies and technologies to operational centers.• Links to cyberinfrastructure. Environmental cyberinfrastructure can be defi ned as a suite of critical enabling tools and research essential to study complex environmental ecosystems (NSF, 2003).Cyberinfrastructure (CI) provides tools for storing, fi nding, analyzing and synthesizing a diverse array of data.It also provides technologies for integrated multi-component modeling systems, visualizations, distributed computations, and adaptive workfl ows.In addition, CI supports the synthesis of observational data and models, provides collaboration tools, and offers new forms of education in science and engineering.It is of paramount importance to link now the development and sustained support of ocean DA research and systems to cyberinfrastructure initiatives.• Technology transfer.An essential aspect of ocean DA research is the eventual transfer of technology to regional and national prediction centers.While in principle such transfers are easy, they are challenging in practice.Issues range from lack of people and funding, to model incompatibilities and unmanageable computational burden.Even though some components of ocean DA, such as data quality control, data streams, and dissemination of results through common technology, are more easily transferable, all technology transfers require extensive and sustained cooperation.VI.Required Infrastructures, Education and FundingCreate strong educational programs and career structures linked to ocean and climate DA, and provide a range of new funding opportunities for theoretical, applied, and operational efforts.• Human resources.Only long-term funding in DA can establish adequate educational and career structures.Most needed items are: (1) sustained programs for students and research groups, (2) intellectual leadership by scientists in the growing DA community, and (3) career scientifi c programmers who are comfortable with evolving complex models, computer languages and cyberinfrastructure. • Computational support.Support is required to sustain educational and operational facilities for ocean and climate DA.In practice, this implies new hardware acquisitions every three years.Computers don't need to be in the same physical location as users, but computer resources must be dedicated to the estimation activity.Because of ever-changing hardware, computer architectures, model codes, and standards, ocean and climate DA activities must have ongoing cooperation with the computational sciences community (e.g., as in the NSF cyberinfrastructure projects).• Required funding.We anticipate that NOPP will continue to be a source of funding for some DA activities, covering both personnel costs and hardware acquisitions.However, new other sustained sources of ocean DA support are urgently needed within the classic ONR and NSF funding programs, as well as within NOAA and NASA.Such support should involve a range of theoretical, applied, and operational possibilities, from small team efforts to larger-group collaborative research.