D6.2: Processing techniques

D6.2: Processing techniques WP6: Implementation of new interfaces and methodologies in coastal systems 30-06-2025/v1.0 1 About this document Title D6.2: Software for ML and DA algorithms Work Package WP6, Implementation of new interfaces and methodologies in coastal systems Lead Partner MET Norway, SHOM Lead Author (Org) Kai H. Christensen (MET.NO) Quentin Jamet (SHOM) Contributing Author(s) Lotta Beyaard (Deltares), Devanshi Pathak (Deltares), Kai H. Christensen (MET Norway), Quentin Jamet (SHOM), Denis Gourves (SHOM), Stephane Raynaud (SHOM), Arthur Capet (RBINS), Florian Ricour (RBINS), Bing Yuan (HEREON), Wei Chen (HEREON,) Jun She (DMI), Vilnis Frishfelds (DMI), Salvatore Causio (CMCC), Giorgia Verri (CMCC), Ivan Federico (CMCC), Julien Brajard (NERSC), Antoine Bernigaud (NERSC), Joanna Staneva (HEREON) Reviewers K. Johnson (Hereon), L. Meszaros (Deltares), Emma Reyes (SOCIB), J. Staneva (Hereon) Due Date 30.06.2025, M18 Submission Date 29.06.2025 Version 1.0 Dissemination Level X PU: Public PP: Restricted to other programme participants (including the Commission) RE: Restricted to a group specified by the consortium (including the Commission) CO: Confidential, only for members of the consortium (including the Commission) FOCCUS: Forecasting and observing the open-to-coastal ocean for Copernicus users is a Research and Innovation action (RIA) funded by the Horizon Europe Work programme topics addressed: HORIZON-CL4-2023-SPACE-01: Strategic autonomy in developing, deploying and using global space based infrastructures, services, applications and data 2023. Start date: 01 January 2024. End date: 31 December 2026. Funded by the European Union (Grant Agreement No. 101133911). Views and opinions expressed are however those of the author(s) only and do not necessarily reflect those of the European Union or the European Health and Digital Executive Agency (HaDEA). Neither the European Union nor the granting authority can be held responsible. 2 / 20 Table of Contents Table of Contents............................................................................................................................ 3 Glossary and Abbreviations.............................................................................................................4 1. Executive summary..................................................................................................................... 6 2. New modelling techniques.......................................................................................................... 6 2.1 Ensemble techniques................................................................................................................. 6 2.2 AI techniques..............................................................................................................................8 3. Optimisation of integrated systems............................................................................................. 9 3.1 Synergistic use of models and observations...............................................................................9 3.2 Coastal-open ocean coupling................................................................................................... 10 4. List of software repositories...................................................................................................... 12 5. References.................................................................................................................................19 3 / 20 Glossary and Abbreviations AI Artificial Intelligence BC Brockmann Consult BGC Biogeochemical CF Climate Forecast (Convention for NetCDF) CMEMS Copernicus Marine Environment Monitoring Service CNR National Research Council (Italy) CRPS Continuous Ranked Probability Score CROCO Coastal and Regional Ocean COmmunity model DCSM-FM Delft3D Coastal Systems Model - Flexible Mesh DPV Data Product Validation DMI Danish Meteorological Institute DCSM Delft3D Coastal Systems Model EO Earth Observation EMODnet European Marine Observation and Data Network ESC Environmental and Societal Challenges FRM Fiducial Reference Measurements FOCCUS Forecasting and observing the open-to-coastal ocean for Copernicus users GLO4ens Global Ocean Physics Ensemble Reanalysis HABs Harmful Algal Blooms HEREON Helmholtz-Zentrum Hereon HR High-Resolution HFR High-Frequency Radar IFREMER French Research Institute for Exploitation of the Sea ISAR Infrared Sea surface temperature Autonomous Radiometer MET.NO Norwegian Meteorological Institute MHWs Marine Heatwaves MHD Marine Hydrographic Directorate (Romania) MSCS Member State Coastal System MSI MultiSpectral Instrument NERSC Nansen Environmental and Remote Sensing Center (Norway) MSCS NetCDF Member State Coastal Systems Network Common Data Form NOC National Oceanography Centre (UK) NW North West ODP Operational Data Product OGCM Ocean Global Circulation Model OLCI Ocean and Land Colour Instrument PDPD Proof of concept Data Product PP Restricted to other programme participants (including the Commission) PU Public PUM Product User Manual RE Restricted to a group specified by the consortium (including the Commission) RIA Research and Innovation Action RMSE Root Mean Square Error ROFI Region Of Freshwater Influence RTK Real-Time Kinematic RBINS Royal Belgian Institute of Natural Sciences S2 Sentinel-2 S3 Sentinel-3 SAR Synthetic Aperture Radar SAV Submerged Aquatic Vegetation SCHISM Semi-implicit Cross-scale Hydroscience Integrated System Model 4 / 20 SDB Satellite-Derived Bathymetry SCHISM SHOM Semi-implicit Cross-scale Hydroscience Integrated System Model Service Hydrographique et Océanographique de la Marine (France) SLA Sea Level Anomaly SOCIB Balearic Islands Coastal Observing and Forecasting System SPM Suspended Particulate Matter SRResNet Super-Resolution RESidual NETwork SST Sea Surface Temperature SWH Significant Wave Height SWOT Surface Water and Ocean Topography TRL Technology Readiness Level TWL Total Water level WMOP Western Mediterranean OPerational forecasting system WP Work Package WW3 WaveWatch III® XBEACH Cross-shore Beach Dynamics Model 5 / 20 1. Executive summary The FOCCUS project (https://foccus-project.eu/) aims to enhance coastal ocean observation and forecasting for Copernicus users by designing and implementing a range of high-resolution data products. This short report is meant to provide a software repository of algorithms, developed under Tasks 6.2 and 6.3, , which involve implementing new modeling techniques (AI and probabilistic approaches) and optimizing integrated systems. It also aims to complement milestones MS6.2 "Nested models implemented in a unified framework" and MS6.3 "New observation operators", as well as deliverable D6.2 "Processing techniques", which is associated with tasks 6.2 and 6.3. The milestones and the deliverable are software and/or data, and demonstrated through codes or data that are accessible via the project space on Zenodo or similar open systems. Brief descriptions of the work are given here to provide context, along with a complete list of the software/data delivered (with links). Not all MCSCs are represented in this work. For a description of the MCSCs we refer to the Section 2 of D6.1 report "Overview of Member State Coastal Systems (MSCS)", which contains a complete list. All software is listed in Table 4.1, and are referred to in the text according to their position in this table. 2. New modelling techniques 2.1 Ensemble techniques Ensemble techniques are used to quantify uncertainties in ocean state forecasts. First initiated in the atmospheric community in the 1990’s (https://www.ecmwf.int/en/about/media-centre/focus/2017/fact-sheet-ensemble-weather-forecasti ng), these techniques have recently emerged in the oceanic community, and this project aims at contributing to this effort. Ensemble techniques consist in producing multiple realizations of the same operational forecast system, each realization only differing in pre-defined metrics of the model (e.g. initial conditions, external forcing, model parameters) depending on the type of uncertainty we aim at quantifying (cf. below). Analyzing the different possible trajectories so produced inform us on possible future ocean states, offering a probabilistic view of the ocean dynamics and prediction errors. Forecast uncertainties are usually clustered in three main types1, accounting for uncertainties in external forcing (e.g. atmospheric conditions, open boundaries and river discharges; Type I), uncertainties in initial conditions (Type II) and uncertainties associated with numerical model imperfections (Type III). The latter two types are subdivided into: Type IIa, associated with small scale, uncorrelated microscopic uncertainties in initial conditions, representing the irreducible component of the non-linear, chaotic dynamics of the ocean; Type IIb associated with spatially organized macroscopic uncertainties in initial conditions, representing our imperfect knowledge of oceanic state at a given time; Type IIIa associated with model inadequacy, reflecting uncertainties in the reproduction of past observation; and Type IIIb associated with model uncertainty, which includes both uncertainties in the parameters values of the various parameterizations used within a given model, and uncertainties across different state-of-the-art models (e.g. ensemble of opportunity). 6 / 20 In FOCCUS subtask 6.2.1, progress on ensemble techniques involve sources of uncertainties classified as Type I, Type IIa, and Type IIIb. These three types of uncertainties have been identified as critical for the different case studies under consideration, including river plumes dynamics, surface drift, wave modelling and wave forecasting. Although these case studies are not all associated with applications detailed in Deliverable D8.1, we envision they will significantly push forward such applications and bring further understanding of potential forecast errors. Note that the ensemble of opportunity approach taken by DMI differs from other approaches as it involves a multi-model strategy while others are applied to a single model. These efforts are summarized below. ● Type I: ○ RBINS: We aim at identifying the respective role of uncertainties in i) atmospheric forcings, ii) land discharges, and iii) boundary conditions in determining the position of river plumes in the southern North Sea. Atmospheric perturbations correspond to the ensemble members of the ECMWF ERA5 ensemble. Land discharges perturbations are reconstructed following a time-series analysis for 85 rivers within our computational domain. A 180-day low pass filter is applied, river per river, to maintain the main trends, while the high-frequency anomalies with respect to this trend are resampled (assuming no temporal correlations). Perturbations of the ocean boundary conditions are obtained by applying the atmospheric perturbations to the larger (North Atlantic) domain used to generate the boundary forcings. The different ensembles will be compared in terms of spread for physical fields (surface temperature, salinity, and current velocities, potential energy anomaly). Following our focus on river plume dynamics, detailed assessments will be based on near-shore surface salinity observations, and aim at characterizing the consistency of the ensembles’ spread using the continuous rank probability score (CRPS). Finally, maps of river plume presence likelihood, enabled by this new ensemble approach will be provided. ○ SHOM: We have interfaced our MSCS with the Global Ocean Physics Ensemble Reanalysis (GLO4ens) reported in Deliverable 6.1. ● Type IIa ○ SHOM: We have completed the implementation of stochastic modelling capabilities in the CROCO modelling platform, where 2D and 3D stochastic fields are generated on the native model grid through auto-regressive processes2. Micro initial conditions ensembles are currently produced thanks to this new modelling capabilities. With these ensembles, we aim at quantifying uncertainties in surface currents associated with ocean turbulence and its feedback on air-sea momentum exchanges, respectively, for application to the case of surface drift (e.g. oil slicks) forecasts. The metrics used to assess the performance of this new modelling technique will thus focus on surface currents magnitude and direction with probabilistic approaches (e.g. ensemble mean and spread, continuous rank probability score (CRPS)). ● Type IIIb: ○ CMCC: We have defined, implemented, and finalized stochastic modelling capabilities for waves at the AdriFS MSCS. Our approach is based on a perturbed ensemble using the air–sea exchange parameter, in order to capture the variability 7 / 20 and uncertainty introduced by the parameterization of the source term inputs in the wave model. The ensemble consists of 12 members, and the outputs include the ensemble mean, selected percentiles, and spread for the variables Significant Wave Height and Mean Wave Period. Model performance will be assessed using metrics such as RMSE, bias, and correlation computed on the ensemble mean of Significant Wave Height, while the Continuous Ranked Probability Score (CRPS) will be used to evaluate the quality of the associated probabilistic distribution. ○ DMI: The purpose of this study is to provide better pan-European Sea wave forecasts than existing products, i.e., CMEMS forecasts and national forecasts. A base study of this work is to validate the existing CMEMS and national wave forecasts, which has been carried out in phase-1. By aggregating CMEMS and national wave forecasts with satellite wave observations, a weighted, multi-model, pan-European Sea real-time wave forecast system has been developed. The preliminary results show improved forecasts compared to the existing ones. However, the weight calculation method needs further optimization, which will be made in phase 2. ○ SHOM: We have completed the implementation of stochastic modelling capabilities in the CROCO modelling platform, where 2D and 3D stochastic fields are generated on the native model grid through auto-regressive processes2. An Air-sea exchange parameter perturbed ensemble is currently produced thanks to this new modelling capabilities. With this ensemble, we aim at quantifying uncertainties in surface currents associated with ocean turbulence and its feedback on air-sea momentum exchanges, respectively, for application to the case of surface drift (e.g. oil slicks) forecasts. The metrics used to assess the performance of this new modelling technique will thus focus on surface currents magnitude and direction with probabilistic approaches (e.g. ensemble mean and spread, continuous rank probability score (CRPS)). 2.2 AI techniques Within the FOCCUS project, advanced AI techniques (including ConvLSTM, CGAN, SRResNet, ensemble SSResNet, MLR) are being developed and applied to enhance ocean forecasting capabilities, particularly in improving: 1) prediction of suspended sediment plumes: ● Deltares has developed a machine learning model that predicts suspended sediment plumes in the Rhine Region of Freshwater Influence, in the southern North Sea. The ConvLSTM machine learning model, which combines a Convolutional Neural Network (CNN) and Long Short-Term Model (LSTM), was used for this. This model structure was optimised, and trained, validated and tested on one year of daily 3D DCSM-FM hydrodynamic output features (water level, currents, salinity, temperature) with a resolution of ~1 km x ~1 km and Suspended Particulate Matter (SPM) observations from the Global Ocean Colour Bio-Geo-Chemical CMEMS product. The SPM output from the 3D DCSM-FM model, with sediment transport processes enabled, was taken as ground truth data. This has resulted in a machine learning model that can predict SPM fields from hydrodynamic data and satellite SPM observations, for 5 days in the future. The algorithm is available on a GitHub repository and through Zenodo (see section 4, software #4). 8 / 20 2) downscaling and super-resolution of ocean fields: ● NERSC has implemented a machine learning approach to downscale DUACS (https://duacs.cls.fr/) Surface height fields. The approach uses a Conditional Generative Adversarial Network (CGAN) and is trained on SWOT data from the first part of 2023. It uses independent SWOT data from the end of 2023 for validation. The super-resolution is done over the Lofoten basin on smaller patches of size 1000 km x 1000 km that are reassembled to reconstruct the full image. The results show that small scales corresponding to SWOT data can be reconstructed. The algorithm is available on a GitHub repository (see section 4, software #7) ● Hereon applied Super-Resolution Residual Network (SRResNet; see section 4, software #1) to downscale the ocean fields including sea surface height and currents in a coastal region of German Bight. High-resolution data from the numerical model SCHISM (Semi-implicit Cross-scale Hydroscience Integrated System Model) and the downsampled data were used for training and testing. The resolution of these ocean fields was enhanced from tens of kilometers to hundreds of meters, with a downscaling factor reaching up to 64. The model also shows good performance when directly using CMEMS data as low-resolution input and high-resolution SCHISM model results as target. Furthermore, an ensemble SRResNet (see section 4, software #2) and multivariate linear regression (MLR; see section 4, software #3) have been applied to do both self-variable (wave to wave) and cross-variable (wind to wave) spatial wave downscaling in the Black Sea. ERA5 data was used as low-resolution input and CMEMS data was used as target. The ensemble method simply averages the predictions from multiple trained epochs, which is generic and easy to be applied to models based on neural networks. This method significantly reduces the prediction instability of the base SRResNet model and the global errors. The ensemble SRResNet performs well in both self-variable and cross-variable spatial downscaling of significant wave height (SWH), while MLR performs well in self-variable downscaling. The development of AI techniques in WP6 have created synergies with other partners in the project, in particular MET Norway with the development of coastal oceanic models driven by AI and SMHI for the downscaling applications. Additionally, AI-based techniques have also been used to implement interface improvements between CMEMS and MSCSs. An example of this is the use of a 4DVarNet-dLSTM model to gap fill SPM satellite observations. These techniques are further described in D6.1. 3. Optimisation of integrated systems The work here is associated with the challenge of obtaining a seamless description of the coastal ocean from the land to the open ocean, and the optimal use of coastal observations. The coastal ocean dynamics is typically associated with strong riverine forcing, with complex water transformation processes and large contrasts in water mass properties over short temporal and spatial time scales. The work in T6.3 is divided into subtasks T6.3.1 synergistic use of models and observations (data assimilation), and T6.3.2 coastal-open ocean coupling (diagnostics). 3.1 Synergistic use of models and observations Work in this task is concentrated in three main regions, each leveraging specific MSCS and observational data for synergistic model-observation use and multi-platform data assimilation: 9 / 20 # Software/ algorithm Deliverable and/or Milestone # Task # and main objective Institution Zenodo descriptor Repository link 16 Observation error analysis D6.2 T6.3.1: Synergistic use of models and observations MET.NO Code related to WPs2 and 6 of the FOCCUS project from MET Norway. Data to the ocean models may be found on https://thredds.met.no/ thredds/catalog.html https://doi.org/10.5281/zenodo.15656700 (cf. https://github.com/metno/FOCCUS/data_a ssimilation/observations/error_statistics) 17 Model diagnostics for freshwater dynamics D6.2 T6.3.2: Land-ocean coupling and diagnostics MET.NO Code related to WPs2 and 6 of the FOCCUS project from MET Norway. Data to the ocean models may be found on https://thredds.met.no/ thredds/catalog.html https://doi.org/10.5281/zenodo.15656700 (cf. https://github.com/metno/FOCCUS/coastal _system/model_diagnostics) 18 Model diagnostics for currents and drift D6.2 T6.3.2: Land-ocean coupling and diagnostics MET.NO Code related to WPs2 and 6 of the FOCCUS project from MET Norway. Data to the ocean models may be found on https://thredds.met.no/ thredds/catalog.html https://doi.org/10.5281/zenodo.15656700 (cf. https://github.com/metno/FOCCUS/appli cations) 19 Observation analysis: Analysis 1hz, 5hz, and 20hz high-resolutio n Sea Surface Height satellite data D6.2 T6.3.1: Synergistic use of models and observations SOCIB https://doi.org/10.5281/zenodo.15748211 /cf. https://gitlab.com/socib/foccus/-/blob/mai n/notebooks/Observation_analysis_stat.ipy nb?ref_type=heads and https://gitlab.com/socib/foccus/-/blob/mai n/notebooks/Observation_analysis_waven umber_power_spectral.ipynb?ref_type=he ads) 16 / 20 # Software/ algorithm Deliverable and/or Milestone # Task # and main objective Institution Zenodo descriptor Repository link 20 Model diagnostics for Sea Surface Height D6.2 T6.3.1:Synergistic use of models and observations SOCIB https://doi.org/10.5281/zenodo.15748211 (cf. https://gitlab.com/socib/foccus/-/blob/mai n/notebooks/Model_diagnostics_for_Sea_S urface_Height.ipynb?ref_type=heads) 21 SHYFEM-MPI model tool for assessment of boundary forcing D6.1/M6.2 T6.1.2: Implementation of new nesting capabilities CMCC This Jupyter notebook presents the Total Kinetic Energy (TKE) from the Child (downscaled) ocean model simulation - computed with SHYFEM-MPI using Daily or Hourly boundary forcing - and compares to the TKE of the Parent model - derived from CMEMS (Copernicus Marine Environment Monitoring Service) Mediterranean Sea Analysis products. https://zenodo.org/records/15638365 22 2D Wave Spectra Forcing D6.1/M6.2 Implementation of CMEMS-MCSC interface improvements HEREON This repository contains a Jupyter Notebook demonstrating how to generate 2D wave spectra inputs for the GCOAST-BS model system (SCHISM-WWM) using data from the Copernicus Marine Environment Monitoring Service (CMEMS). This repository is related to the FOCCUS project, specifically Deliverable D6.1, and aims to https://doi.org/10.5281/zenodo.15690198 17 / 20 # Software/ algorithm Deliverable and/or Milestone # Task # and main objective Institution Zenodo descriptor Repository link provide a standalone demonstration of the methods used in this deliverable. 23 Multi-model ensemble wave prediction D6.2 T6.2.1 Ensemble technique DMI Pan-European wave aggregation https://doi.org/10.5281/zenodo.15696952 24 Nudging CMEMS T/S to improve initial condition and forecast D6.1/M6.2 T6.1.2: Implementation of new nesting capabilities DMI First release https://doi.org/10.5281/zenodo.15696216 25 SST bias correction through Nudging D6.1 T6.1.2 Implementation of new nesting capabilities RBINS The repository contains the North-Sea model setup, Fortran additions for SST nudging in COHERENS, and associated compile/run scripts. The source code of the model COHERENS v3.0 is available at : https://gitlab.naturalsciences.be/e comod/coherens_v3. https://doi.org/10.5281/zenodo.15706316 26 MOHID-SWAN operational coupling D6.1 T6 2.1 Ensemble technique +ATLANTIC Implementation of a one-way coupling system between the hydrodynamic model MOHID and https://doi.org/10.5281/zenodo.15756032 18 / 20 # Software/ algorithm Deliverable and/or Milestone # Task # and main objective Institution Zenodo descriptor Repository link the wave model SWAN. 19 / 20 5. References [1] Stainforth, D. A., Allen, M. R., Tredger, E. R., & Smith, L. A. (2007). Confidence, uncertainty and decision-support relevance in climate predictions. Philosophical Transactions of the Royal Society A: Mathematical, Physical and Engineering Sciences, 365(1857), 2145-2161. [2] Brankart, J. M., Candille, G., Garnier, F., Calone, C., Melet, A., Bouttier, P. A., ... & Verron, J. (2015). A generic approach to explicit simulation of uncertainty in the NEMO ocean model. Geoscientific Model Development, 8(5), 1285-1297. [3] Bonaduce, A., & Raj, R. P. (2025). Sea-level and currents (Versión 1) [Data set]. Zenodo. https://doi.org/10.5281/zenodo.14825347 [4] Vergara, O., & Pujol, M.-I. (2025). Sea Surface Height for coastal applications obtained from Level-3 satellite altimetry (V1.0) [Data set]. Zenodo. https://doi.org/10.5281/zenodo.14938132 [5] Yuan, B., Ricker, M., Chen, W., Jacob, B., Pham, N.T., Staneva, J. (2025): Statistical spatial downscaling of significant wave height in a regional sea from the global ERA5 dataset. Ocean Engineering 329, 121100. https://doi.org/10.1016/j.oceaneng.2025.121100 [6] Yuan, B., Jacob, B., Chen, W,, & Staneva, J. (2024): Downscaling sea surface height and currents in coastal regions using convolutional neural network. Applied Ocean Research, Vol 151, 104153, https://doi.org/10.1016/j.apor.2024.104153 20 / 20