IV. WOCE OCEAN MODELLING

IV.1. Overview of the status of ocean models

 Realistic simulation of the ocean's dynamical structures and properties is the goal of ocean modelling in WOCE. It is an essential step in improving coupled ocean-atmosphere modelling for climate prediction.

 Many aspects of ocean modelling have been supported by WOCE, with an emphasis on basin-scale and global simulations, increasing resolution and improving model physics (see for example, WOCE Strategy for Ocean Modelling, 1993). The resulting progress is impressive, and ocean-only models have become increasingly realistic in recent years and are now used with 20-60 vertical levels, and a horizontal resolutions as fine as 1/6 of a degree. They aim to resolve mesoscale dynamics generated by internal instabilities of the ocean and vorticity dynamics of boundary currents. These large-scale, eddy resolving models require very significant computational resources and manpower, some using an entire super computer for months at a time.

 Computationally intensive climate studies (including coupled ocean-atmosphere studies) which require long term integration of ocean models are still performed at coarse resolution, as they are still not possible at eddy-resolving scale. However, errors in the highly parameterised coarse resolution model simulations are probably amplified by the long integration times that are required for climate studies. This problem has been addressed by new parameterisations which have been developed during WOCE, and which have permitted significant progress. As a result, ocean models with slightly higher resolution than commonly used now in climate studies (i.e. near 1 degree), using the newly developed parameterisations of eddy-induced mixing and topographic stress, can in the near future be expected to address some of the long-standing deficiencies of coupled ocean climate models.

 The number of available ocean models has increased during WOCE, with the development of alternative models to the well established ones. One notes encouraging examples such as the emergence of isopycnic models (Bleck and Boudra, 1986), and the application of topography-following vertical coordinates to basin scale circulation studies. These may be considered major contributions to model development, as they were positive and creative efforts which enhanced the ocean model diversity and thus also the need for model inter-comparison studies. While these are encouraging developments, the number of available ocean models lags behind the number of different atmospheric models, - the existence of many models facilitates investigation of the statistical distribution of model solutions.

 During the WOCE AIMS period, the established and new ocean models should be further tested, new parameterisations should be developed for coarser models, and much model development work should occur for high resolution models. The recommendations given here represent the WOCE effort to define the needs of ocean climate modelling in order to advance WOCE's goal of developing ocean models for climate prediction.

 IV.1.1. Coarse resolution models

 Coarse resolution ocean circulation models used in climate studies typically have a resolution of the order of a few hundred kilometres. Used as diagnostic tools, they qualitatively portray the large scale structure of oceanic gyres and the gross features of the thermohaline circulation. They give a picture of the ocean which approximates that derived from historical hydrographic observations.

 However, these coarse models are highly parameterised in order to represent the geometry, bathymetry and dynamics not explicitly resolved. The need for such parameterisations, together with the limitations of presently available parameterisations, is one of the major reasons for model deficiencies in the description of the oceanic circulation and variability, and limits the predictive skills of coarse models.

 Progress has been made in the parameterisation of mesoscale eddy-induced mixing (e.g., Gent and McWilliams, 1990) which have been shown to improve several deficiencies of the coarse resolution simulations (Danabasoglu and McWilliams, 1995).

 The treatment of the bottom topography effects has also been investigated. Parameterisations of the eddy-induced bottom stress have drawn some attention (Holloway, 1992). The use of a topography-following coordinate (sigma-coordinate) in the vertical, commonly used in atmospheric models, but previously limited in oceanography to process studies (Beckmann and Haidvogel, 1993) or to coastal or regional applications (Ezer and Mellor, 1992) has recently been applied to general circulation models with encouraging results.

 However, major deficiencies still remain in coarse resolution model representation of ocean mixing, convection and subduction processes, in the description of structure and strength of boundary currents and flows through narrow sills, and in the estimation of meridional heat and fresh water transports. Furthermore, model sensitivity, and the picture models give of the ocean variability are strongly dependent on surface flux conditions. Consequently, it is not yet demonstrated that these coarse models have predictive skills for long term climate studies.

 One of the important goals of WOCE modelling is the development of improved coarse resolution ocean models for climate studies.

 IV.1.2. High resolution (eddy permitting/resolving) models

 These models resolve much of the along-isopycnal mixing due to eddies, better represent the vorticity dynamics of western boundary currents, and better resolve important topographic features than do the coarse models. Due to resource constraints, they are normally used in ocean-only simulations, with a crude or no parameterisation of the atmospheric feedbacks. They appear sufficiently advanced to be useful for interpreting fields measurements, and for studying various aspects of the ocean physics. They can also provide information useful to the planning of field programmes.

 However the computational cost and manpower requirements of these high resolution model experiments create several difficulties. First, experiments with these models cannot be carried out for very long periods, and model solutions actually produced are therefore not equilibrated and thus not independent of the initial conditions. Second, there are serious difficulties in assessing the synoptic scales produced by these model experiments because of a lack of observational data on the appropriate spatial and temporal scales. In this regard, satellite altimetry data are very useful. Subsurface float data will provide some clues on the variability at depth, but they are still too sparse, and most floats do not yet provide all necessary state variables.

 A further problem resulting from the large cost of running high resolution ocean models is that a very limited number of modelling groups are carrying on such model integrations (3 or 4 in the US, two in the UK, one in Germany, and one in France - see Table IV.1), often with limited manpower and a shortage of engineering and technical support. Large and organised modelling efforts must be planned in the remaining phase of WOCE if the goal of modelling climate is to be achieved. (See section IV.4.2)

 In spite of the impressive progress of high resolution models in recent years, there are still deficiencies that need to be investigated and corrected. Their representation of the eddy field is not quantitatively accurate, and simulated eddy kinetic energy levels are lower than observed in many parts of the ocean. They do not represent accurately enough aspects of the oceanic variability such as frequency spectra, or ENSO events. Model solutions still strongly depend on parameterisations used to represent small scale mixing, convection and mixed layer dynamics. The search for an adequate three-dimensional resolution to correctly resolve the topographic constraints, baroclinic instability, and the vorticity dynamics of boundary currents is still in progress. It is likely that major improvements to high resolution model solutions will be the outcome of this search.

 IV.1.3. Major unresolved issues in ocean modelling

 McWilliams (1996) recently summarised the main issues in ocean modelling as "customary and alternative dynamical approximations of the fundamental fluid mechanics, the parameterisation of essential processes that occur on spatial and temporal scales smaller than can be resolved in model calculations, the boundary and initial conditions, the domain geometry, and the numerical algorithms". It is clear that some of these are issues of general applicability, some are more specific to coarse resolution models, and some are relevant only to eddy-resolving models.

 An open issue of general importance is the atmospheric forcing of the ocean. Theoretical progress over the past few years significantly enhanced our understanding of the sensitivity of the thermohaline circulation in ocean models to the parameterisation of the air-sea interaction, in particular the parameterisation of air-sea exchanges of heat and fresh water. In parallel, there have been considerable improvements in the determination of the surface winds and surface heat and fresh water fluxes over the ocean from observations. However, analyses from NWP Centres, from satellite data, or from recent VOS climatological estimates have not yet reached the accuracy required for long term climate studies (below a few W/m2). We recommend that the effort to compare and improve all sources of air-sea flux estimates should be continued. The need for continuous coverage of air sea fluxes during the WOCE period may, however, be partially satisfied by the reanalyses performed by the main NWP Centres. Still, simple yet accurate parameterisations of air-sea fluxes are needed for ocean-only model studies.

 A second issue of general importance is the data set against which the models are compared and the methods of comparison. Coarse resolution models are usually compared to hydrographic estimates of a mean state of the ocean. As indicated above (section II.2.1), there is a question about what state of the ocean is represented by the WOCE data set given interannual and seasonal variability. Tracer can be utilised to improve the estimate of the "mean". At seasonal time scales, the scarcity of appropriate data results in a much greater problem with model/data comparison, mainly at mid to high latitudes, but also in the monsoon regions. Data comparison is especially difficult for eddy-resolving models given the lack of eddy-resolving observations.

 A third issue of importance is model resolution. The oceanic general circulation, as opposed to that of the atmosphere, strongly depends on the dynamics of narrow boundary currents. The minimum resolution needed to adequately represent these currents is still unclear, and therefore, the resolution issue in both coarse and eddy-resolving models requires further work. In that respect, the needed horizontal and vertical resolutions are not independent, nor the resolution and the parameterisation of sub-grid scale processes. The relation among these elements is still not well understood.

 Since relatively coarse resolution models are likely to be used in climate studies in the near future, it is also urgent to quantify the differences between high and coarse resolution models, and to develop new parameterisations for the coarse models (eddy mixing, diapycnal mixing, surface mixed layers, topography, narrow straits, etc.).

 IV.2. WOCE-related ocean modelling

 IV.2.1. Existing and planned efforts

 An attempt has been made to summarise existing activities in the field of WOCE modelling and assimilation (Table IV.1). It has not been possible to obtain information from some countries, and information is incomplete from others. However, the existing evidence shows that there are about 250 principal investigators (not including Post-docs and students) at work on the problem at present. Direct modelling activities range from single-basin studies with coarse resolution models to global eddy-permitting models. Assimilation studies include methods ranging from simple nudging to Kalman filters, adjoint methods and representers.

 While these numbers seem encouraging, it must be emphasised that only a very small portion (estimated at 10%) of these ocean modellers are involved in model development work. The rest are using existing models for various simulations and assimilation purposes. The small number of model developers is less than optimum given the many open issues and challenges confronting us in the area of ocean model development. This needed model development work is quite technical and has high manpower demands in the form of technical assistants (post-docs, and engineers).

 Funding agencies in the various countries are urged to dedicate more resources to ocean model development. More specific recommendations follow in the sections below.

 IV.2.2. Available and needed resources

 Researchers have access to several ocean model codes, including:

 MOM/MOMA/Cox/OCCAM/POCM/...
LODYC
SPEM, POM and other sigma-coordinates
MICOM and other layered models
HOPE
eddy-resolving quasi-geostrophic
large-scale geostrophic
non-, or almost non-hydrostatic
other versions under development
new approaches (finite element/volume, other numeric schemes, etc.)

 This set of codes has grown during the period of WOCE. It still represents a smaller number of codes than is available in meteorology, and still leaves room for improvement in terms of needed new parameterisations etc. Thus more work is needed in the area of model improvement and parameterisation development.

 Supercomputing, as Table IV.1 shows, varies in its provision between countries. Frequently it forms a shared resource with other research areas. In the UK, for example, environmental researchers have direct access to 15% of an MPP machine and 29% of a vector machine. Both these shares are fully utilised now. Twenty runs of a five-year global assimilation on a platform four times faster than that available would take two years of environments share of the next MPP machine.

 Data storage is generally not seen as a problem. Using figures for a 1/4 of a degree resolution global model (a basin at 1/6 to 1/12 is similar) with 36 levels, a restart dump plus assimilation information would require 2.6 Gb of storage. An assimilation over 5 years, assimilating once every two weeks, involves a total storage of order 1/2 Tb. Discs of this size are now available and will become cheaper.

 Data transmission may be a problem in countries with slow networks. The 1/2 Tb would take a day to transfer across a typical country's network without other users, so that physical movement of the data itself, in the form of digital linear tapes, may well be more efficient. This holds even more true when local or international networks are considered, which are frequently far slower than national.

 Last but not least is the issue of manpower requirements. This is a resource that needs to be supplemented in many countries, as mentioned in the previous subsection.

 To fully carry out the WOCE AIMS phase heavy (and often extra) computing commitments from national funding agencies will be required.

 IV.3. Strategy for model development

 IV.3.1. Towards the WOCE objective: developing a model for climate prediction

 WOCE's main objective is to develop an ocean model for climate prediction. Climate prediction must be obtained using coupled ocean-atmosphere-ice climate models, and thus WOCE's objective must be the development of an ocean GCM that may be used within a coupled model. The large computational requirements of such coupled models put serious constraints on the resolution of the ocean component that may be used in coupled models during the next few years. This large computational cost involves both running the coupled model itself, and also running the atmospheric and oceanic components separately to steady state in order to initialise the coupled model, and in order to calculate the commonly used flux correction terms. Presently, coupled models are making the transition from 4- degree resolution ocean models to 2- degree resolution. It is likely that this resolution will increase to better than 1 degree within the next 5 years.

 It is thus important that the ocean modelling community within WOCE works on the development of the next generation ocean models to be used in coupled models. It is clear that even a 1/2 degree resolution is not sufficient to accurately represent many oceanic processes such as boundary currents and eddy kinetic energy. It is the responsibility of WOCE modellers to demonstrate what higher resolution accomplishes for coupled model studies and climate prediction purposes.

 Work on the development of ocean models for climate studies should proceed on three parallel tracks. First, there must be work on high resolution ocean-only models (order of 1/12 degree) that provide as realistic simulation of the oceans as is presently possible. Second, process models should be developed that may eventually provide parameterisations in climate models. Finally medium resolution, improved models that are of higher resolution than are used today in coupled models, yet that are not forbiddingly expensive computationally should be developed. These medium resolution models will serve as the next generation ocean models within coupled climate models.

 The high resolution models and the process models, together with the WOCE data, can serve to test the medium resolution models that are intended for use within coupled models. These medium resolution models will be tested by comparing key elements of their solutions with the more detailed models. For example, the meridional transports may be compared with those of the high resolution models, and mixing rates or strait transports may be compared with the relevant process models.

 By carrying out such comparisons of the medium (say 1/2 degree) resolution models with the more detailed ocean models and WOCE data on the one hand, and with the coarse models presently used for climate prediction on the other hand, WOCE modellers need to demonstrate that better ocean models are indeed needed for climate prediction purposes. This WOCE modelling work should prepare the next generation ocean model for coupled climate model studies, a task that is not expected to be attempted by those running coupled models. Within five years, as computer power increases, WOCE should be able to hand to the coupled model community an eddy-permitting, medium resolution (1/2 degree) ocean model that is the best that can be used in coupled models and that fits the computer resources available then.

 IV.3.2. Process models and parameterisations

 All existing GCMs have inherent problems. These problems are fewer and less crucial than at the start of WOCE because of the predicted increase in computing power; global or quasi-global models at a resolution of 1/6 of a degree are now practical. For the time scale of WOCE AIMS, however, two sets of problems will remain. One is relevant to a specific model code; the other set is generic to all existing codes.

 Level models (MOM, Cox, OCCAM, LODYC, HOPE, etc.) have persistent difficulties with step-like topography, which causes vertical mixing and signal loss in deep overflows. The standard centred difference advective scheme is noisy; noise remains a difficulty with both B- and C-grid formulations. On a B- grid, potential vorticity is not conserved, and this can generate spurious instabilities. Layer models (MICOM, etc.) have difficulties relating to the definition of density, thermal wind, and retreat of mixed layers in springtime; this latter also produces noisy fields. Sigma-coordinate models require great care in the treatment of the potentially erroneous horizontal pressure gradient, especially in equatorial regions.

 All models describe the small-scale (both unresolved and just resolved) poorly, especially near ocean boundaries, so that process studies will be needed to improve them. The SMWG encourages work in the following areas, inter alia: Flow through sills, straits and gaps in a rotating system involves narrow ageostrophic layers which carry an important part of the through-flow, as well as interactions with the turbulent bottom boundary layer and areas of strong vertical turbulent mixing of tracers (cf. the Deep Basin experiment). Where models, even with embedded finer grids, are unable to reproduce these features accurately then process modelling will be necessary in order to produce physically based parameterisations.

 Few large scale ocean models possess turbulent bottom boundary layers (Killworth and Edwards, 1997). Bottom layers are capable of moving heat, salt and tracers through very large distances in ocean basins while maintaining a strong signal, and are thus efficient transporters of heat and salt, processes which are not included at present in models. Process modelling of streamtube models is the way forward, and attempts are being made at two-dimensional layers. The manner in which fluid is detrained from turbulent layers back into the interior laminar fluid is not well understood, and both process and numerical models will be necessary to include the phenomenon. While there are several surface mixed layer models available, despite much work in the last decade, none exists that is applicable globally; the development of such a model is an urgent need for the AIMS phase.

 Save for basin calculations, true eddy-resolving (as distinct from eddy-permitting) calculations remain too expensive for most researchers. When such computations are made, trade-offs on storage and cpu can involve coarsening the vertical dimension so that the resolution requirements for baroclinic instability are broken (the 1/12 MICOM N. Atlantic run being an example). Parameterisations of eddy effects remain therefore a crucial issue.

 All the above features require parameterisations based on physical understanding. The most urgent is an eddy parameterisation, since many of the AIMS calculations will be made with non-eddy-permitting models. Two types of parameterisations have recently been proposed: one involving the redistribution of potential energy (the Gent and McWilliams (1990) scheme) and one involving statistical eddy dynamics, which adjusts the kinetic energy (the Holloway (1992) Neptune scheme). Practitioners report some success with both schemes, with water masses better reproduced using the GM scheme, and flows closer to reality with the Neptune scheme. It is clear that some users have adopted GM because it relieves the Veronis effect near ocean margins (indeed, it has been used in eddy-permitting models!). It should be noted that the removal of the Veronis effect (enhanced mixing in the presence of steeply sloping isopycnals and consequent unrealistic vertical motions) seems to be achievable by simply smoothing the topography and reducing the horizontal background mixing used in tensor diffusivity parameterisations.

 There remain many difficulties with both schemes: the need for closure of Gent/McWilliams at boundaries, which is rather ad hoc, and the equally ad hoc selection of length scales for the Neptune scheme.

 The examination and testing of existing parameterisations, the production of new parameterisations, and a critical inter-comparison of coarse models using such schemes with eddy-permitting or -resolving models is viewed as an urgent priority.

 IV.3.3. Numerics

 Traditional (centred-difference) numerical schemes remain popular in GCMs because of their conservation properties. There is considerable evidence that improved properties can be obtained with modified schemes. Finite element/volume methods, semi-Lagrangian methods, shape-preserving advection schemes, automatic grid embedding, multigrid methods, etc., are just a few of the approaches either in use or under experimentation. Clearly many of these will reach fruition too late for the AIMS phase, but some (e.g., some advection schemes) are in place and robust now.

 The development of novel numerical techniques for ocean models is a priority area currently with few active participants. This topic needs to be encouraged and supplemented with additional researchers. The development of new numerical methods and interaction with other areas of computational continuum dynamics is also strongly encouraged.

 IV.3.4. Model-data comparison

 Model testing and comparison with observations should proceed along several lines.

 First, many practical modelling questions concern ignoring or parameterising processes lost when resolution is degraded to make feasible long runs. For these questions, simulation results from high resolution models can be used to evaluate coarser models for scales that the coarser models are intended to resolve. For this reason some short calculations with the best models run at very high resolution are necessary.

 The first and most direct model-data comparison is between model results and selected raw observations. All WOCE data described in Section III can be used for this purpose. Imperfections in temporal and spatial sampling in the WOCE datasets however can cause discrepancies with even a perfect model. Although WOCE has produced a significant number of observations, the available data are still sparse, approaching model spatial resolutions only at a few sites or along a few lines that were occupied occasionally. This makes comparisons intrinsically probabilistic because no model can be expected to simulate the effects of all processes that occur between the observations. A major part of the WOCE observations primarily address the climate of the ocean, rather than its detailed evolution, and comparisons should largely focus on comparisons of climates. In this area some tracers offer the opportunity of a 'mean state snapshot' and should be more widely used, as well as the smoothed large-scale circulation derived from the one-time survey. However, data from satellite altimeters XBT and surface drifter programmes do provide significant information about temporal variability. Modellers must find ways to use these data for model evaluation and improvement, in this way fulfilling WOCE objectives.

 One approach is comparison of observed and model climatologies with particular focus on the tests of key processes (advection, lateral and diapycnal mixing, subduction) imposed by budgets over scales that are addressed by the WOCE observations. This approach puts heavy emphasis on a dialogue between modellers and observers over how discrepancies might best be resolved. An advantage of the approach is that the apparent disagreements are clearly drawn and understood, as are their resolution. The disadvantage is that the hypotheses implicit in treating the data and making the comparison are likely not to be as clearly stated or easily tested as they are in true assimilation analyses.

 With the progress in data assimilation during the WOCE AIMS phase, products from data assimilation can be used to prepare data sets for use in testing model performance. The result of assimilating data into an ocean model is a better estimate of the ocean state than model simulation alone. This requires, of course, that the model used for the assimilation is of a sufficient resolution and accuracy to prevent aliasing the assimilation result towards a model state which is wrong. Criteria for model performance and model error estimation need to be developed and refined to aid in deciding whether a given model is capable of producing meaningful assimilation results. Then assimilation products can serve as easily-used comparison fields for testing models. Further, data assimilation results from the "best" model can serve as a benchmark for other models. As models improve, data assimilation results can be reproduced using progressively improved models to provide new benchmarks for further model improvements.

 The processes of data assimilation and model testing can be combined within a single assimilation exercise configured to define substantive discrepancies between the data and model. This approach has the advantages that explicit statements must be made of the assumptions used to resolve discrepancies (the descriptions of likely data variability and model errors) and clear statements about the likelihood that the model and data are consistent are obtained. In practice the validity of these assumptions and their unknown impact on the conclusion leave room for debate on the formally objective conclusions but the debate, itself, sharpens the questions that must be answered before complete comparisons of models and data can be carried out. Disadvantages are significant computational cost and difficulty of the computations.

 IV.3.5. Long term objectives

 The previous discussion of WOCE modelling objectives has been in terms of models run at current resolution. Finer resolution will almost certainly become feasible in the future, and WOCE modelling will almost certainly exploit the advantage. Nevertheless it is worth stating the long-term objective.

 We should aspire, by the year 2007, to have a global ocean model resolution high enough to yield highly credible representations of the flow of boundary currents over complex topography, and the flow of deep currents through narrow gaps. It would resolve several baroclinic modes of the deformation radius, and therefore the model might give very different eddy fluxes from a model that only resolves the first. It would also yield values for testing parameterisation schemes. Indeed, a small-basin early version would suffice for that purpose. Such a 'super-model' would need about 2-4 Terabytes of RAM. The CM-5 Connection Machines at NRL and LANL now have 32 Gigabytes, so the requirement might well be met in a decade from now.

 IV.4. Modelling recommendations

 IV.4.1. Ocean modelling workshop

 Several useful workshops on modelling as well as on modelling and assimilation have been held so far as part of WOCE. After discussions among the CLIVAR Numerical Experimentation Group (2) , the SMWG and the WOCE Scientific Steering Group, an ad hoc group was formed with the main objective of examining ways in which advances in both observations and modelling made during WOCE could aid the development of climate models. This group is structuring an ocean modelling workshop, to be held in Boulder (August 10-13, 1998).

 Three central topics have emerged for the workshop;-

 1) The representation of ocean dynamics in climate studies, including ocean processes critical for the formation of deep and intermediate waters, meridional overturning and heat transport, and its response to atmospheric forcing.

 2) Assessment of the accuracy and limitations of the ocean models used for climate studies, defining the fields that are important in the ocean response to forcing and that need to be reproduced well.

 3) Identify a set of oceanic fields to be used for model testing, leading to better co-ordination, feedback and progress for model development.

 IV.4.2. Joint development of community models

 Ocean model development requires heavy investments in intellectual effort, manpower and computational resources. It is often a too heavy mission for a single researcher or a small modelling group. Community efforts in ocean model development, such as the WOCE Community Modelling Effort (CME), have existed in the past and should be planned for the near future.

 The US has taken the lead in this respect and at a meeting in Boulder in April 1997 which was organised by US WOCE but included participants from other major modelling projects a Community Consortium for Ocean Modelling (CCOM) was agreed. The CCOM would

 a) address large-scale scientific problems, such as those associated with NAO, ENSO/PNA, and global change forcing , and

 b) adopt a comprehensive (multi-disciplinary, whole system) modelling approach.

 Specific goals of the project would be to co-ordinate efforts to develop and refine ocean models.., co-ordinate experiments.., to provide benchmarks.., provide infrastructure..., develop products for forcing, initialisation and evaluation..., promote co-operation between modellers and computer scientists and engineers..., create testbeds for innovative algorithms...

 Evidently the US modelling community had been alerted that resources for community effort would succeed only if an effort was made which combined both physical and biogeochemical interests. They have responded to this as seen in the report of the meeting (Boulder 1997).

 The US effort in setting up a cross-project Community Consortium for Ocean Modelling is applauded. Other National Committees are encouraged to consider if similar action is appropriate within national or trans-national jurisdictions.

 IV.4.3. Resource requirements for AIMS modelling

 The following is a rough estimate of the resources needed to address WOCE's modelling objectives over the next few years. We assume that analysis is made of 20 ocean basins, which can include global state estimates, etc. We further assume 5 prime-movers per basin, giving 100 workers over a period of 5 years. Assuming an equal number of support staff, and that it is these latter which need non-institutional support, approximate modelling costs for the AIMS phase are:

 100 staff @$60k/yr for 5 yr $30M
200 workstations @ $30k + renew after 3 yr $6M
Infrastructure (DACs, computer lines, etc.) $10M ?
Students - how many? ?
Supercomputing:
5 million T3D processor hours @ $750/256 processor hours $15M
TOTAL $61M

 This figure is around 6% of the observational phase of WOCE and could probably be provided through existing national funding methods.

 An important component of the resources needed in WOCE AIMS are scientists to work on ocean modelling and model development; consequently it is important to pursue programmes that encourage young researchers to move into modelling activities via, for example post-doc programmes.

 IV.4.4. Models and Model products

 The first goal of WOCE calls for producing improved ocean models for climate prediction purposes. It is envisioned that at the end of WOCE AIMS one or multiple versions of high resolution eddy permitting/eddy resolving ocean models (1/4 -1/12 degree) will be available for eventual use in coupled ocean-atmosphere models for climate simulation and prediction.

 In addition, coarse resolution ocean models (1-1/2 degree horizontal resolution) and process models or parameterisation modules which deal with specific oceanic physical processes such as mixing and convection will also be provided by WOCE. These coarse resolution and process models may provide immediate benefit to climate modellers to improve their practical coarse resolution coupled models for short term climate prediction and long term climate simulations. However, we must clearly identify, quantify and emphasise the shortcomings of the coarse ocean models (and the high resolution models too) to the climate community. These shortcomings can result in misrepresenting or completely omitting important physical processes in the ocean which may be significant for long term climate variability.

 Access to model output is essential for joint analysis of data sets and model results, and for comparison of models with data in order to facilitate further development of the models. It is recommended that the ocean modelling workshop committee and the Data Products Committee devise a plan for recommending types of useful model output and for linking modelling sites together. Support will be required for this activity.

 TABLE IV.1. Computing resources for ocean modelling, as of mid-1997

Country       Institute      Prime-   Activities                   Computing/codes
                             movers

Australia     CSIRO Oceanog  6        Inverse, Coupled, Assim,     MOM, SPEM
                                      Parameterisation 

              Ant. Coop Res. 2        Coupled, Ice, Eddy dyn,      HOPE

              CSIRO Atm      4        Coupled                      MOM

              Bureau Met     3        Assim, Coupled 

              Univs          8        Global,                      MOM, SPEM
                                      Spectral/processes, Model
                                      development 

Israel        Weizmann       1        Coupled, Adjoint             MOM 

S. Africa     U. Cape Town   1        Indian O., non-eddy 

Japan         CSSR (Univ)    4        Global, Coupled, Assim       MOM+tracers

              Kyoto          3        Assim, Deep Conv             Nonhydro

              Hokkaido       3        Ocean-ice, Assim 

              MRI            5        SubAnt Subduction, Coupled 

France        Univs.and Labs 15       High Res Atl (N,S,Whole)     SPEM, LODYC
(combined     Brest,Grenoble          coupled/forced       
effort)       Paris,Toulouse          4-D Assim, Kalman filter,    SPEM,MICOM
              through                 Adjoint                      LODYC,IMG-QG
              CNRS,CNE
              IFREMER,                Coarse Res, Coupled Global   LODYC
              SHOM, ORSTOM            Inverse Atlantic

Germany       IFM            6        NA & Ind Highres, Processes  MOM, SPEM
                                      Atl Coupled Medres, Ice

              AWI            8        Atl Highres, Global Medres   MOM, SPEM
                                      Ice, Polar Processes         HOPE

              Hamburg        10       Coupled, Coarse global       HOPE, large-scale
                                                                   geostr

UK            SOC            12       Global eddy res + Assim,     MOMA (MPP), MICOM
                                      Global non-eddy,
                                      Pre-assim/assim, Inverse,
                                      Model develop &
                                      application, Remote sensing 

              UEA            3        Eddy res                     MOMA 

              Edinburgh      2        Assim                        MOMA 

              Met            10       FOAM, Global coarse,         MOM
                                      coupled 

              ECMWF          5        Coupled, Assim               various

              Reading        2        Coupled, Processes           MOMA 

Canada        Dalhousie      3        Adj, season cyc, TOPEX/POS   Large-scale 
                                                                   geostr 
                             4        Adjoint shelf dyn            Various 

Spain         ULPGC          5        Tides, waves, QG, process    Local codes

              UPC            2        S. Ocean, mesoscale          3D fint element

Netherlands   IMAU           5        Stabilty, global circ,       MOM, local,
                                      variability, assimilation    C90,T3E,SP2 

Other Europe  ?              6        Coarse global                MOM 

Russia        Shirshov       2        Eddy-perm, initialisation    MOM
                             3        Adjoint, steady              Large-scale
                                                                   geostr

              Dept of Comp   2        Adjoint regional SST         MOM 
              Math Acad Sci 

              Marine Hydro   2        Reduced Kalman               MOM
              Inst 

Norway        DNMI           4        Nordic non-eddy, Nordic      POM,MICOM,OSMOM
                                      eddy-perm, process, Assim,   Y-MP, T3D 
                                      climate 

              IMR            3        Nordic non-eddy, Skagerrak   NORWECOM/POM
                                      eddy-perm, Barents 
                                      non-eddy biology

              Nansen         2        Assim (ensemble Kalman       MICOM
                                      filt), Nordic non-eddy,
                                      Carbon 

              NP             2        Barents/Kara process         SPEM/SCRUM/MICOM

USA           Univs thru NSF 20       Model development            MOM,MICOM, SPEM

              Univs thru ONR 20       Bias to coastal

              NCAR           5        Coupled                      MOM 

              GFDL           3        Climate/coupled              MOM 

              PMEL/AOML      2        Regional                     MOM

              NCEP           5        Operational prediction       MOM 

              LANL           3        Model development/MPP        POP, MICOM 

              MIT            5        Regional/global;satellite    SPEM, MOM, MIT
                                      model development;
                                      assim with adjoint 

              NRL-Stennis    6        Global/regional for          NLOM
                                      operational

              Naval Postgr.  4        Global ocean & applications  POCM, POP

              Goddard        6        Satellite, ocean/ice         MOM 

              JPL            3        Satellite, Kalman            POP, Red-grav

              OSU            5        Assimn with repres,          many
                                      Kalman, tides, steady                        
                                      estimation

              FSU            3        Adjoint                      Red-grav           

TOTAL                        ~250
Return to Table of Contents