Objectives

COMEDI’s main objective is to modernise and harmonise data assimilation practices across all Copernicus Marine Monitoring and Forecasting Centres, so that they can use new technologies and new types of ocean observations to achieve better accuracy and reliability.

Four specific objectives

The seven Monitoring and Forecasting Centres surround four shared objectives.

Arctic Monitoring and Forecasting CentreARCMFC Baltic Sea Monitoring and Forecasting CentreBALMFC Black Sea Monitoring and Forecasting CentreBLKMFC Mediterranean Monitoring and Forecasting CentreMEDMFC Iberia-Biscay-Ireland Monitoring and Forecasting CentreIBIMFC Global Ocean Monitoring and Forecasting CentreGLOMFC North-West Shelf Monitoring and Forecasting Centre: North-West ShelfNWSMFC 1. Advance data assimilation 2. Make best use of observations 3. Improve interoperability 4. Prepare for operational use Modernise&Harmonise

Select a segment to go to that objective

1

SO-1Advance data assimilation methods

Gaussian (EnKF) non-Gaussian forecast error

Advance the data assimilation methods in current Monitoring and Forecasting Centre operations by implementing novel techniques for uncertainty representation and assimilation, integrating AI, to achieve better accuracy and reliability of the operational products.

Today in the Monitoring and Forecasting Centres → with COMEDI

TopicTodayCOMEDI
MultiscaleMostly optimised for a single scale; the North-West Shelf centre uses two scalesMultiscale covariances and scale-dependent localisation in three-dimensional variational (3DVar) methods and ensemble Kalman filters (EnKFs)
EnsemblesMostly static covariances; only the Arctic and Global centres run dynamic ensemblesDynamic ensembles in EnKFs, blended into 3DVar
Non-GaussianAll operational methods assume Gaussian errorsThree non-Gaussian filters for coupled biogeochemistry and sea-ice data assimilation
AIPrototypes exist but are not used in operationsAI bias correction, super-resolution and generative AI as a core data assimilation method

Key results

  • 2 multiscale methods for uncertainty representation
  • 1 perturbation scheme for coupled physics–biogeochemistry models
  • 2 hybrid climatological–dynamical covariance techniques
  • 3 AI methods for systematic errors, subgrid uncertainty and ensemble diversity
  • 3 non-Gaussian ensemble filters
  • 2 generative AI algorithms used as core data assimilation

Work packages

WP2Forecast uncertainty representationWP3Core assimilation method
2

SO-2Make best use of observations

CIMR Sentinel-3 NG sea-ice Tb wide swath Argo float

Prepare the Monitoring and Forecasting Centres and users to take maximum benefit from novel ocean observations, upgrading the data assimilation methods to make better use of observation information and implementing diagnostic tools to evaluate their impact systematically.

Today in the Monitoring and Forecasting Centres → with COMEDI

TopicTodayCOMEDI
Sea iceRetrieved sea-ice and surface temperature productsDirect assimilation of brightness temperatures in coupled ocean–sea ice data assimilation
High resolutionObservation errors treated as uncorrelated, except along-track altimetryCorrelated observation errors in both variational methods and ensemble Kalman filters
New missionsPrototype operators for Sentinel-3 Next Generation (S3-NG), the Copernicus Imaging Microwave Radiometer (CIMR) and ocean colour reflectanceS3-NG wide-swath altimetry, CIMR brightness temperatures and reflectance assimilated in real Monitoring and Forecasting Centre settings
ImpactData-denial experiments; evaluation across Monitoring and Forecasting Centres is hardAssessment protocol, systematic observing system simulation experiments (OSSEs) and observing system experiments (OSEs), and forecast sensitivity to observation impact (FSOI) for variational and ensemble methods

Key results

  • 1 updated observation operator for the CIMR mission
  • 1 method for correlated observation errors in wide-swath altimetry
  • 2 tools for forecast sensitivity to observation impact (FSOI)
  • An AI emulator of FSOI at least 10× faster, in the Global Monitoring and Forecasting Centre system
  • 2 OSSEs: Sentinel-3 Next Generation in the North-West Shelf Monitoring and Forecasting Centre and CIMR in the Arctic Monitoring and Forecasting Centre
  • OSEs for reflectance data in 3 Monitoring and Forecasting Centres

Work packages

WP4Observation operator and uncertaintyWP5Observation impact evaluation
3

SO-3Improve interoperability

GLONWSBLKMEDBALIBIARCCIGARNEMOVAROceanVarOAKE-Med-BGCPDAFMROANEDASCOMEDI innovations as standalone modulesinterface between softwarestandalone module

Improve the interoperability of the Monitoring and Forecasting Centres by developing modular software components and shared interfaces, enabling a more efficient exchange of new methods among operational centres. There is no single common system: each Monitoring and Forecasting Centre keeps its own data assimilation software, and methods travel through interfaces between software packages and through standalone modules that several packages can use.

Today in the Monitoring and Forecasting Centres → with COMEDI

TopicTodayCOMEDI
Data assimilation softwareEach Monitoring and Forecasting Centre develops its own data assimilation software, and methods rarely pass between centresEach Monitoring and Forecasting Centre keeps its own software; new interfaces let packages use each other's methods, such as MROA and NEDAS using the core of the Parallel Data Assimilation Framework (PDAF)
New methodsBuilt deep into centre-specific code and costly to transferReleased as standalone modules that several data assimilation packages can adopt
AILittle connection to modern AI toolsInterfaces to AI libraries in Python and Julia, shared between partners

Key results

  • 7 developments from work packages 2–5 implemented on the data assimilation platforms of at least two Monitoring and Forecasting Centres
  • 3 interfaces that give other data assimilation software access to the core methods of the Parallel Data Assimilation Framework (PDAF), linking 4 ensemble-based Monitoring and Forecasting Centres
  • 2 variational platforms sharing the same NEMO observation operator
  • 5 data assimilation software packages interfaced to AI models
  • The STOPACK stochastic perturbation package available for both the NEMO and HYCOM ocean models

See where each innovation is shared, and its status →

Work package

WP6Interoperable software development
4

SO-4Prepare innovations for operational use

2 workshopsDemosUsersARCNWSGLOIBIBLKBALMEDNavigation, shippingSearch and rescueOil spillsWater qualityFisheries, foodClimate

Prepare data assimilation method innovations for operational use by demonstrating them in realistic Monitoring and Forecasting Centre configurations, reaching Technology Readiness Level 6, and by engaging Copernicus Marine Service stakeholders for feedback.

Demonstrations in the seven Monitoring and Forecasting Centres

  • ARCArcticNew non-Gaussian and generative-AI filters; direct assimilation of brightness temperatures from the AMSR2 satellite radiometerSea-ice extent and drift forecasts for navigation and shipping
  • BALBaltic SeaParticle flow filter and the local Kalman nonlinear ensemble transform filter (LKNETF) in coupled physics–biogeochemistryBottom salinity, which drives oxygen conditions
  • BLKBlack SeaCorrelated observation errors for SWOT (Surface Water and Ocean Topography mission) wide-swath altimetry; generative AI in coupled physics–biogeochemistryDepth of the oxycline; circulation for oil-spill response
  • GLOGlobal OceanMultiscale and AI ensemble methods with SWOT altimetry; ocean colour reflectance assimilation; AI bias correction in the CIGAR reanalysisSurface currents, carbon fluxes, ocean heat content and the Atlantic Meridional Overturning Circulation (AMOC) for climate and food security
  • IBIIberia-Biscay-IrelandMultiscale and AI ensemble methods with SWOT altimetrySurface currents checked against drifters
  • MEDMediterranean SeaAI-assisted hybrid covariances; high-order ensemble Kalman filter, autotuning and AI super-resolution for biogeochemistryEnvironmental predictors for species distribution models
  • NWSNorth-West ShelfCorrelated observation errors for SWOT altimetry; full ensemble with STOPACK; multiscale hybrid covariances; Sentinel-3 Next Generation wide swathSurface velocities for drift modelling and search and rescue

Engaging users

Two stakeholder workshops, at mid-term and near the end of the project, bring users and the Advisory Board (Copernicus Marine champion users, national services and related EU projects) together with the teams running the demonstrations. Their feedback shapes the demonstrations and the roadmap into operations.

Work packages

WP5Observation impact evaluationWP7Demonstration in Monitoring and Forecasting CentresWP8Dissemination, exploitation and communication