| 2 |
< < < l l l 8 < , " , ! R $ < < < 6 R < < P 4 < < < < g l l P I < $ $ < Revised Work Statement, ECCO-GODAE January 2004
Overall.
This project is directed at the full exploitation of new capabilities for combining global scale observations with the most powerful available ocean circulation models. The results are essential for a variety of important uses, including seasonal to interannual forecasting of the climate system, of numerous national contributions to the IPCC process for assessing climate change and its impacts (including sealevel change and carbon uptake), and the much firmer scientific understanding of the ocean circulation and its variability in time. The latter ultimately influences such diverse problems as national fisheries and environmental policies, national security (in its widest sense), and even energy policy.
The work will be a collaboration of several institutions, an arrangement necessary given the complexity of the problem. This complexity occurs because of the need to have available the latest, most sophisticated and highly efficient ocean circulation numerical models, the understanding of the rich diversity of ocean data types--suitably quality controlled--and ready access to optimization software suitable for problems of dimension many orders of magnitude larger than previously attempted. The work builds directly on what has been accomplished by the Estimating the Climate and Circulation of the Ocean (ECCO) Consortium with NOPP support over the past four years, and aims to extend and exploit the capability for the wider community. To that end, the Consortium (SIO, JPL, MIT) is being widened to include GFDL, AER, NCEP and GSFC. The ECCO results can be obtained through extensive published papers and reports, available online at http://www.ecco-group.org.
The major components of the proposed work will include: (1) the production of seasonal to interannual forecasts, their testing and comparison; (2) the continued estimation of dynamically consistent, global and regional day-to-day estimates of the three dimensional time-evolving ocean circulation at the highest feasible resolution; (3) the public distribution of all products including both the scientific and application communities; (4) continued evolution of models and methodologies so that the estimates continue to reflect the state-of-the-art; (5) a quality-controlled global data stream reflecting the great majority of the oceanic observations now available.
Work Plan for Global Ocean State Estimation (Lead: MIT)
The overall goal of the work is to produce on-going best-estimates of the time-dependent, absolute ocean circulation employing as much of the global ocean data sets as is practicable. These are intended for public dissemination, but will be of interest primarily to the broad scientific community. These estimates, which we have been making as part of the ECCO Consortium for the past 4 years, will continue to improve with both new data and the use of better model and optimization codes. It is also our intention to increase the model resolution as resources permit, and to employ the results ourselves for certain scientific purposes, the latter in collaboration with both our ECCO-GODAE partners and with other organizations. The existing ECCO system is used to carry out global 1 degree estimation beginning in 1992 and running to the near-present, employing the data sets described in Stammer et al. (2002a,b,c). The estimates are based upon the method of Lagrange multipliers (adjoint or 4dvar in other communities). This entire software package, involving general circulation model, data sets, and estimation algorithms, has already been ported to GFDL and run sufficiently to confirm that the computing resources are adequate for our needs at this resolution.
In the first year of this work, the MIT group (Heimbach, Wunsch, a research scientist to be appointed, and the resident modeling group (A. Adcroft and others)) will work primarily with the SIO state estimation system that is now being replicated on the GFDL machines. The calculations will be carried forward, closer to the present, and we will establish a means by which periodic updates will be done and publicly posted. The primary computations will be done at GFDL using specially modified ECCO-queues and allocations of machine resources. The major contact person at GFDL, who will help solve technical problems and work with the results, will be Dr. Shaoqing Zhang. The MIT group will work with Zhang and others at GFDL in comparisons of the ECCO-model products with those derived from the ongoing construction of MOM-code based systems. The impact of ARGO float and other new data (the new GRACE geoids, time-dependent bottom pressure from GRACE, etc.) will be evaluated.
The second year is anticipated to be the time in which we start producing both high resolution regional products (North Atlantic, Southern Ocean) and global results with a 1/4 degree system. A central issue is whether eddy permitting models significantly increase the skill of the estimates compared to coarser resolution models with eddy parameterizations. A decision on the production model can only be made after experiments are run. In the second and later years, we anticipate shifting some of the production work to the NCAR climate system computers. This shift has the advantage of both freeing the GFDL computers for more experimental work, and using resources funded by another agency which has expressed interest in seeing the facility used in this way. (Note that under the existing NOPP program, the 50+ year state estimates (from 1949) are being carried out at NCAR. This work would continue with the collaboration of D. Stammer at U. of Hamburg.) The scientific applications of primary interest to the MIT group overlap the GFDL-IPCC goals, including global biogeochemical cycles, earth rotation, heat budget and sealevel changes. These calculations, because of their direct interest to the GFDL scientists, will remain primarily at GFDL. The system, both regional and global, will be available for use by CLIVAR investigators both through MIT and GFDL.
The MIT collaboration with JPL and GFDL will occur in several different ways. S. Griffies (GFDL) and A. Adcroft (MIT) will be working toward a collaborative MIT-GFDL framework for climate models. The objective is to create a pool of software that will be used by both modeling groups to replace otherwise redundant code. The exercise is mutually beneficial; the MIT group gains by adopting the GFDL- supported infrastructure tools, such as the diagnostics manager, and the GFDL group gains access to new numerics (e.g. momentum discretizations of vector invariant equations on a C grid), new formulations (e.g. non-Boussinesq and non-hydrostatic) and code that is more readily adjointable (via the TAF compiler). The initial task is exploratory, establishing to what degree model software is already compatible and redundant between the two groups and to agree to target designs toward which each group will evolve. Coordinated adoption of programming structures and conventions used in both models will enable interoperability of the lower level numerical routines, parameterizations and tools. It is anticipated that the majority of software handling scalars (thermodynamics, tracers and sub-grid scale parameterizations) can be readily interoperable with relatively minor structural changes to the two models. Adoption of ESMF by both models during this process will facilitate interoperability of auxiliary software such as diagnostics and data management.
The MIT group will take primary responsibility for oceanographic data flow and quality control. These data are of mutual interest to all consortium institutions and will also be public. (As one example of the required activity, the known regional bias offsets between TOPEX/POSEIDON and Jason-1 data are leading to work to decide how best to form a combined data set without generating discontinuities in altimetric constraints on the model.)
Carbon cycle and possibly decadal forecasts and initialization of global change scenarios, will be a common scientific focus of both MIT and GFDL groups. The JPL collaboration will occur through several different elements, including a common interest in obtaining realistic low-latitude estimates, in the technical elements of adjointing MOM-4, and through the common collaboration with E. Tziperman at Harvard (who is active in this area, but not a formal part of the proposal). The latter includes work with P. Heimbach on the use of singular vectors obtained from the adjoint system for study of initial condition perturbations. G. Gebbie, an MIT graduate student working with Heimbach and Wunsch on the MIT-ECCO model, will, on completion of his thesis, take up a position with Tziperman, providing another direct link between the groups.
In addition, D. Menemenlis of JPL will work closely with the MIT group on the development of the high-latitude and ice modeling components which are major elements of the intended MIT products. With the use of the cubed-sphere geometry, the global model can now include the full Arctic. The ice model, which is essential to any high latitude computation, has been developed at JPL, and the collaboration which will involve further model development, analysis of the observations and the ultimate inclusion of ice into the optimization calculations, will directly involve the JPL group. The most immediate application will, however, be to the MIT component of the collaboration.
The MIT group will develop a next-generation ECCO state estimation system, incorporating many modifications and improvements that have become possible since ECCO began. These modifications and improvements are expected to include the cubed-sphere coordinate system, removal of the Boussinesq approximation, incorporation of atmospheric pressure loading, inclusion of topography in the state vector, coupling to the more capable ice model, and numerous other potential changes. The latter include, for example, self-attraction and seafloor loading---properties of importance in using time-variable gravity data. Also to be examined in collaboration with JPL is the inclusion in the state estimation system of a hybrid of the Lagrange multiplier (adjoint) approach with the filter/smoother sequential system, to take advantage of their complementary nature.
The standard ECCO production runs will continue, but will gradually shift to a "mark-2" ECCO system based upon the above changes, carefully compared to the existing ("mark-1") version. Both would become public. This system conversion is a reasonable goal for year 3, but would continue into the out-years.
R. Ponte and AER colleagues will be scientific collaborators on this work with C. Wunsch, with a view toward commercializing the products if possible, but also in analyzing the scientific results (a major effort) and to help oversee the ongoing calculations.
Work-plan for Climate Modeling (Lead: GFDL)
An ECCO-like adjoint state estimation system and its associated tangent-linear and adjoint models will be established based on MOM4. The target ocean component will be the OM3 model developed at GFDL for climate change scenario experiments. This model has a resolution of 1 degree with equatorial enhancement to 1/3 degree and 50 vertical levels with 20 levels in the upper 200 meters. The model has been coupled to an atmospheric one, and the experience with such coupled systems is projected to be a necessary element for seasonal-interannual forecasting, in any future extension of ECCO-GODAE, and for the wider climate change community. Ultimately, the MIT-ECCO model will be similarly coupled, but some years of experience and collaboration with GFDL will probably be required before it could be used in a production mode. OM3 parameterizations include Gent-McWilliams eddy fluxes and the KPP vertical mixing scheme. Development of the tangent linear and adjoint models will rely heavily on the use of the TAF compiler developed by FastOpt---and used in ECCO---for automatic tangent linear and adjoint code generation. The project can be broadly separated into 3 main stages:
Phase I will terminate with the completion of OM3, pending assessment of its viability for climate scenario integrations. Two candidate models currently exist and a final determination will be made prior to December 2003 (GFDL).
Phase II will involve modifications to the ocean model code for compatibility with TAF along with improvements to TAF in support of f90-compliant code in MOM4. This work is already underway in collaboration with E. Tziperman at Harvard. Subsequent work will include: modifications to existing algorithms in MOM4 for TLM/Adjoint accuracy; TAF directives for performance; minimization algorithms; and parallelization, and involving the GFDL, Harvard, JPL and MIT groups.
Phase III is the implementation of an "ECCO-like" system using OM3.
A version of the ECCO system is currently running at GFDL. The performance of the objective function minimization for the ECCO system will be analyzed by Dr. Shaoqing Zhang at GFDL. Data handling and quality control for in-situ and satellite data in use for ECCO will be merged with existing systems in use at GFDL (GFDL). A 1-degree version of the ECCO model will be maintained at GFDL and kept up-to-date during the course of the collaboration (MIT). Similarly, OM3 will be maintained in near real-time using more traditional OI techniques to assimilate similar datasets. As with the other elements of this work, data will be made accessible through OpenDap (web access) and LAS (web graphical interface) portals to facilitate comparison.
Through a longer-term focus directed towards improved automatic differentiation capability via TAF, it is anticipated that tangent linear and adjoint models will be a standard practice in future ocean model development. This development will allow for improvements in model physics to be easily evaluated in the context of adjoint sensitivity. In this light, the implementation of an "ECCO-like" system at GFDL will take place in the broader context of how ocean (and ocean biogeochemical) models are developed and interpreted, in addition to more direct applications such as state estimation and forecast initialization.
Workplan for Seasonal-to-Interannual Climate Forecasting (Lead: JPL)
This element is aimed primarily at exploiting the ECCO capability for seasonal climate forecasting. We aim to develop a next-generation operational seasonal climate forecasting system by implementing advanced data assimilation methods and by employing additional data types not yet used in the operational systems. For this goal, the project brings together expertise in climate modeling and forecasting (NCEP, GFDL, GSFC), data assimilation (JPL, MIT, NCEP, GSFC), and observing systems (NCEP, SIO, JPL).
Specifically, the ECCO ocean state estimation methods will be implemented in the NCEP seasonal climate prediction system and GSFCs Poseidon Ocean model. In this component, the methodology consists of a Kalman filter, a smoother, and the adjoint model. The system's rigorous estimation procedures permit 1) optimal use of existing observations, 2) incorporation of new observations, 3) production of physically consistent estimates suitable for process studies, and 4) sensitivity analyses of physics/dynamics and of observing systems.
For several years, NCEP has maintained a quasi-operational forecast system focused on the Pacific Ocean and a forecast model that coupled the model to a global atmosphere. NCEP is currently engaged in upgrading this entire system, and a new global system became operational in September 2003. This new Global Ocean Data Assimilation System (GODAS) is based on MOM3, developed at GFDL. The model has a one degree resolution overall, with a one-third degree meridional resolution in the tropics, similar in resolution and in model complexity to those of the present ECCO model.
Like the old Pacific Ocean system, GODAS uses an objective mapping (3DVAR) method for state estimation. NCEP has employed this method for its simplicity and low demand on resources that are critical requirements in an operational environment. Once implemented, the ECCO Kalman filter requires no more resources, and is thus attractive from an operational standpoint.
Following GODAS's operational transition (i.e., when the system operation has transferred from NCEP's Environmental Modeling Center to NCEP Central Operations), NCEP will begin developing the next generation GODAS based on MOM4. The present investigation will constitute part of this development effort in integrating new data sets, improving data and background errors, and implementing and evaluating new estimation methods.
The ECCO system is characterized by its goal of statistical optimality. The rigorous mathematical formulation of the Kalman filter is also useful for diverse observation types---because the method requires that the models compute only the equivalent of the observations. Smoothing yields estimates with closed budgets of heat and other properties (physical consistency) that are invaluable for scientific analyses of the assimilation products. The adjoint provides an effective means to assess the model's sensitivity to various controls, thus permitting insight into mechanisms of climate variability and the impact of different observing systems on the accuracy of the estimate.
GSFC/NSIPP has led NASA's effort to exploit satellite observations to improve forecasting seasonal-to-interannual climate variability. GSFC is also a center of the Consortium for Ocean Data Assimilation for Seasonal-to-Interannual Prediction (ODASI). ODASI is focused on improving ocean data assimilation methods and their implementation in support of forecasts with coupled general circulation models. Other co-investigators at NCEP and GFDL are also part of ODASI, and together with GSFC, constitute half the ODASI consortium. GSFC will provide NASA expertise in seasonal forecasting and will also serve as the focal point to leverage ODASI investigations that will, in turn, contribute to the consortium's goal.
The present project's work plan consists of two inter-connected investigations; one concerning the Kalman filter/smoother and the other regarding the adjoint model.
The ECCO Kalman filter/smoother will be implemented with the MOM4 ocean model as it is configured at NCEP (see also the above discussion of the Climate Modeling workplan), and the Poseidon ocean model employed in GSFCs analysis system, so as to improve the forecasting accuracies of the respective systems. The lead investigators in this effort are I.Fukumori (JPL), D.Behringer (NCEP), and M.Rienecker (GSFC). The development of the Kalman filter/smoother will incorporate latest advances in error source modeling and estimation, including sources of diabatic model error as well as time- and spatially-correlated errors. See Detailed Breakdown of Seasonal-to-Interannual Forecasting Work for further explanation of the implementation plan. As part of this effort, the Kalman filter/smoother will also be used to produce a MOM4-based near-real time analysis, to replace the system presently being maintained at JPL using the MIT model. Such a system synthesizes data from operational observational systems such as ARGO and TOGA-TAO in near real time to support GODAE-related research and applications.
In the context of seasonal-to-interannual climate forecasting, the adjoint of the MOM4 ocean model will be developed principally as a tool to analyze the sensitivity of the forward model. As a secondary objective, the adjoint will be employed in experiments to improve the Kalman filter/smoother assimilation and, separately, to establish an ECCO/SIO-like climate analysis system using MOM4 at GFDL. This investigation will be led by T.Lee (JPL), A.Rosati (GFDL), and C.Wunsch (MIT) with collaboration by the whole group plus that at Harvard. The main tasks involved in this development include (see also the Climate Modeling workplan above): Adjoint code parallelization, including related design of re-computation, memory storage, and I/O directives for the adjoint compiler. Modules for cost function and control variables. Testing (e.g., gradient check by comparison with finite-difference sensitivity). Minimization module, assimilation (e.g., over progressive 6-month intervals). Initialization of seasonal climate forecast.
Combined Annual Goals:
Year 1
Begin global state estimation at GFDL using SIO-type methods (MIT/GFDL)
Begin testing of cubed-sphere and other improvements (ice model) in global system (MIT/GFDL/JPL)
Produce a routinely accessible global data stream (MIT/GFDL)
OM3 development (GFDL)
Begin a 20+ year reanalysis using GFDL's OI data assimilation maintained in near real-time (1 month lag). (GFDL)
Tangent linear/adjoint model development for MOM4 (GFDL/Harvard/JPL/MIT).
Improvements and extensions to TAF (JPL/MIT/GFDL/Harvard) including application to MOM4 and ESMF.
Continued model development with emphasis on parameterizations such as GM,KPP and keeping them adjointable. Improvements in computational efficiency (GFDL/Harvard/JPL/MIT).
Incorporation of new data types (MIT/JPL).
Establish the elements of GODAS system (model/data system/methodology) (NCEP/JPL).
Upgrade NCEP ocean model to MOM4 (NCEP) and begin simulations.
MOM4 NCEP ocean model and GSFCs Poseidon model ported to JPL (JPL/NCEP/GSFC).
Comparisons of seasonal-to-interannual forecasting model results and data, including model error calibration (JPL/NCEP/GSFC).
Derivation of Kalman filter and smoother of MOM4 and Poseidon (JPL/NCEP/GSFC).
Exploration of commercial interest in products (AER).
Year 2
Shift of global production toward mark 2 system (MIT/GFDL). Continuation of global estimates. Begin regional estimates in conjunction with Clivar investigators.
Optimization of ECCO and GFDL "ECCO-like" models for applications to seasonal-to-interannual forecasting, decadal initialization and ocean carbon studies. (GFDL/JPL/MIT).
Experiments with OM3 system (JPL/GFDL/NCEP/GSFC) for seasonal-interannual forecasting, and comparison with other systems
Experiments with new data types/constraints (JPL/MIT/SIO); comparisons with independent data types (e.g., earth polar motion).
Kalman filter and smoother assimilation with MOM4 and Poseidon (JPL/NCEP/GSFC).
Adjoint sensitivity analyses of the circulation to various controls and observing systems (JPL/GFDL/NCEP).
Adjoint data assimilation refinement of Kalman filter/smoother estimates (JPL/MIT).
Year 3
Shift of global production state estimates of Mark 2 system (MIT/GFDL) and continued production of global estimates. Continued work with Clivar regional and global investigations.
S-I forecast/hindcast experiments with coupled models and different methods. (JPL/GFDL/NCEP/GSFC), and comparisons with withheld data.
Evaluation of various models/methodologies for transfer of best system to operational agencies (JPL/NCEP/GSFC/GFDL).
A near-real time analysis system based on MOM4 and Kalman filter/smoother will be implemented to utilize operational observations from ARGO and TOGA-TAO to support GODAE activities (JPL/NCEP).
Years 4, 5
We do not provide specific goals for years four and five, as it is too difficult to extrapolate with any accuracy so far into the future. The ECCO-GODAE group however, would maintain the global state estimates, be producing working regional estimates for various CLIVAR groups, and be improving the SI forecasting systems. By years four and five, we anticipate that the operational elements within NCEP will be routinely using the major ECCO and ECCO/GFDL software. We anticipate that the MIT/GFDL combined modeling effort will have produced considerable convergence in the software---work that will also continue. Overall goals include the ability to use coupled atmosphere/ocean/ice models in a practical way for state estimation, no later than the end of year 5.
Appendix. Detailed Breakdown of Seasonal-to-Interannual Forecasting Work (JPL/GFDL)
Year 1:
The main focus of the first year will be in establishing the elements of the new GODAS system. These elements include the ocean model, data system, and components of the assimilation system.
Behringer (NCEP) will lead in upgrading NCEP's ocean model to MOM4. Various forward simulations will be carried out with this model to assess the model's fidelity and to calibrate its accuracies. In particular, to gauge the range space of the ocean model in simulating climate variability, the model will be run in the coupled configuration with the global atmosphere that is used in climate forecasting. The simulations will be compared with observations so as to estimate data and model errors.
The identical MOM4 NCEP ocean model will also be transferred to JPL (Fukumori and Lee) to carry out some of these simulation experiments and to perform the error calibration based on the method of so-called covariance matching. The GSFC Poseidon ocean model (Rienecker) will also be employed in this error calibration process. Inclusion of the GSFC model and its assimilation system will permit additional means to assess mutual model errors and to help identify efficient means to represent the errors.
The covariance-matched error estimates (Fukumori, lead) will be assessed and compared against error models utilized in the NCEP 3DVAR system and in the GSFC OI system. Elements of the covariance matched estimates will be employed in the NCEP (Behringer) and GSFC (Rienecker) assimilation systems to assess the fidelity of the error estimates vis-a-vis those native to the respective systems. Alternatively, the 3DVAR and OI error estimates will be used as prior estimates in covariance matching to further optimize the latter estimate.
New data sets will also be integrated into the existing operational suite of observations. These include Argo temperature and salinity profiles. Behringer (NCEP) and Roemmich (SIO) will lead in the utilization of Argo data. A quality control algorithm similar to that for the present NCEP temperature analysis will be devised and used for the entire project. Other observing systems will be incorporated in later years.
The ECCO assimilation system will be ported to MOM4 and Poseidon following the model set ups. Fukumori and Rienecker will lead in deriving elements of the Kalman filter for the NCEP and GSFC models. Model error sources presently modeled by the ECCO system will initially be employed along with its basic state reduction schemes. Analysis of the two models will help further design an advanced state reduction scheme. In particular, diabatic errors will be analyzed by comparison with in situ profiles of temperature (e.g., TAO and Argo) and will be incorporated in the state reduction schemes. Time-mean errors (biases) will be estimated by inverting the time-mean state of the model using absolute measurements such as temperature.
Lee (JPL) and Rosati (GFDL) will begin deriving the adjoint of MOM4 in collaboration with Tziperman (Harvard) and TAF developers (FastOpt) plus Heimbach at MIT. Funding for the first two will be provided by additional resources external to the present investigation. The MOM4 adjoint development will build upon the GFDL/Harvard prototype adjoint for the vector version of the model and on the ECCO experience in deriving the adjoint for the parallel MIT model. Re-computation and memory storage will be analyzed and I/O directives will be implemented for the adjoint compiler (TAF).
Year 2:
In Year 2, the ingredient elements of the assimilation system derived in Year 1 will be combined and ocean data assimilation experiments will be initiated and its results analyzed.
Fukumori (JPL) will lead in deriving the state error covariance matrices employing the asymptotic approximation using the doubling algorithm. In collaboration with Behringer and Rienecker, these errors and the resultant filters will be examined vis-a-vis the NCEP 3DVAR weights and the GSFC OI operators. Elements of the filter (e.g., structure of underlying prior covariance) will also be employed in the 3DVAR (Behringer) and OI (Rienecker) systems for testing.
The filter and smoother will be implemented with the ocean models and ocean data assimilation experiments will be carried out. The fidelity of these estimates will be examined with independent observations and with the native 3DVAR and OI assimilation estimates. Identical data streams will be employed in the assimilation. Individual observing systems will be employed to assess the relative impact of the observations and to examine the skill of the assimilation system relative to different observations. Metrics established by ODASI will be used to assess the skill of the analyses in comparison to NCEP and GSFC estimates and others of the ODASI consortium. These analyses and comparisons will be conducted jointly by JPL, NCEP, GSFC, and GFDL.
New observing systems will continuously be explored. In particular, Fukumori will lead in the utilization of GRACE measurements. These include constraints on time-mean sea level in combination with satellite altimetry and estimates of changes in ocean bottom pressure. The combination of sea level, bottom pressure, and temperature and salinity profiles (Argo), allows distinguishing eustatic, thermosteric, and halosteric effects on sea level that are important in ENSO forecasting.
The skill of the model and assimilations will also be measured with respect to other geodetic observations, such as length-of-day and polar motion, which provide unique integral measures of the skill of the ocean models. The GRACE measurements will also be introduced into the NCEP and GSFC analyses for studying and for constraining the operational assimilation.
The adjoint model will be tested through various gradient checks. Adjoint sensitivity will be compared with sensitivity computed by finite-amplitude perturbations. Lee will lead in conducting these sensitivity experiments. Modules for cost function and control variables will be established in collaboration with S.Zhang (GFDL) and E. Tziperman of Harvard. The adjoint model will first be employed to evaluate ocean circulation and to assess the sensitivity of the circulation to various controls, including sensitivity to aspects of the observing system. The adjoint will also be introduced to NCEP as a tool to help analyze its assimilation products.
Zhang will also evaluate and implement minimization algorithms, building upon those of the ECCO system. The adjoint model together with the minimization routines will be used to re-analyze estimates of the Kalman filter by expanding the control space (e.g., to include uncertainty associated with mixing coefficients). Lee will lead this effort, and will examine the optimization in 6-month intervals to minimize the computational requirements. The adjoint model will also be employed to implement an ECCO/SIO-like climate analysis system at GFDL, assimilating all extant observations, but extending the analysis period to present. Zhang and Rosati will lead this analysis at GFDL. The Kalman filter estimates and its error covariances will be employed in both adjoint assimilations so as to optimize the adjoint calculation's convergence and to capitalize on the explicit cross covariance estimates of the Kalman filter. Combining the assimilation schemes will be jointly pursued by MIT, GFDL, and JPL.
Year 3:
In year 3, the ECCO-like assimilation system will be employed in hindcast experiments using the coupled forecast models so as to evaluate the impact of the new initialization scheme. Behringer and Rienecker will lead in conducting the forecast experiments, while Fukumori will oversee the assimilation. Metrics established by ODASI will again be used to assess the skill of the predictions and the impact of the assimilation. These metrics include the near surface heat content of the tropical Pacific and daily averaged temperature and current velocity measured by the TAO/Triton/PIRATA arrays. A near-real time analysis system based on MOM4, like the one currently maintained at JPL based on the MIT model, will be implemented to synthesize data from operational systems such ARGO and TOGA-TAO to support GODAE-related applications.
Computational resource requirements of the assimilation system will be assessed and optimized for an operational environment. Behringer, Rienecker, and Fukumori will evaluate the ECCO Kalman filter/smoother in light of these requirements and the skill of the initialization with respect to existing analysis systems. Depending on the results of these analyses, suitable elements of the new initialization scheme will be incorporated in the respective operational systems.
Experimental forecasts will also be carried out using the adjoint-based assimilation with the NCEP model. The skill of the forecast will be analyzed in comparison to other initializations so as to assess deficiencies in respective assimilation schemes. Deficiencies will in turn be analyzed to improve the respective analyses. Behringer will be responsible for conducting the forecast computations, while Lee and Fukumori will take the lead in assimilation and the evaluation of the results.
The ECCO/SIO-like adjoint assimilation system will be established at GFDL permitting periodic extension of the ECCO/SIO climate analysis (with MIT and GFDL). In particular, an analysis extending from 1992 to present (2006+) will be obtained using an expanded suite of observations including those employed in seasonal climate forecasting of NCEP and GSFC. In particular, these will include Argo floats, GRACE measurements, and up-to-date satellite altimeter observations and sea surface temperature measurements. GFDL and MIT will take the lead in this extended period analysis with JPL contributing to the technical aspects of adjoint modeling.
Here, MIT means the combination of MIT and AER.
PAGE
PAGE 13
< q 0 1 > S> @ @ @ @ UV mV nV uV W W T[ [[ ^ ^ 0a ;a ;d d d d >r Gr G H z { h 0J mH nH u |
Parent Directory
| 
Revision Log
| 
Revision Graph