Gabriele Accarino

Accurate ocean forecasting is essential for supporting a wide range of marine applications. Recent advances in artificial intelligence have highlighted the potential of data-driven models to outperform traditional numerical approaches, particularly in atmospheric weather forecasting. However, extending these methods to ocean systems remains challenging due to their inherently slower dynamics and complex boundary conditions. In this work, we present MedFormer, a fully data-driven deep learning model specifically designed for medium-range ocean forecasting in the Mediterranean Sea. MedFormer is based on a U-Net architecture augmented with 3D attention mechanisms and operates at a high horizontal resolution of 1/24°. The model is trained on 20 years of daily ocean reanalysis data and fine-tuned with high-resolution operational analyses. It generates 9-day forecasts using an autoregressive strategy. The model leverages both historical ocean states and atmospheric forcings, making it well-suited for operational use. We benchmark MedFormer against the state-of-the-art Mediterranean Forecasting System (MedFS), developed at Euro-Mediterranean Center on Climate Change (CMCC), using both analysis data and independent observations. The forecast skills, evaluated with the Root Mean Squared Difference and the Anomaly Correlation Coefficient, indicate that MedFormer consistently outperforms MedFS across key 3D ocean variables. These findings underscore the potential of data-driven approaches like MedFormer to complement, or even surpass, traditional numerical ocean forecasting systems in both accuracy and computational efficiency.

Precise and reliable climate projections are required for climate adaptation and mitigation, but Earth system models still exhibit great uncertainties. Several approaches have been developed to reduce the spread of climate projections and feedbacks, yet those methods cannot capture the nonlinear complexity inherent in the climate system. Using a Transfer Learning approach, we show that Machine Learning can be used to optimally leverage and merge the knowledge gained from global temperature maps simulated by Earth system models and observed in the historical period to reduce the spread of global surface air temperature fields projected in the 21st century. We reach an uncertainty reduction of more than 50% with respect to state-of-the-art approaches while giving evidence that our method provides improved regional temperature patterns together with narrower projections uncertainty, urgently required for climate adaptation.

Several scientific fields, including climate science, have undergone radical changes in the last years due to the increase in the data volumes and the emergence of data science and Machine Learning (ML) approaches. In this context, providing fast data access and analytics has become of paramount importance. The data space concept has emerged to address some of the key challenges and support scientific communities towards a more sustainable and FAIR use of data.The ENES Data Space (EDS) represents a domain-specific example of this concept for the climate community developed under the umbrella of the European Open Science Cloud (EOSC) initiative by the European Commission. EDS provides an open, scalable, cloud-enabled data science environment for climate data analysis on top of the EOSC Compute Platform made available through a user-friendly JupyterLab GUI. The service integrates into a single environment climate datasets from well-known initiatives (e.g., CMIP6), compute resources, data science and ML tools. It was launched in the context of the EGI-ACE project, it is accessible through the EOSC Catalogue and Marketplace (https://marketplace.eosc-portal.eu/services/enes-data-space) and it also provides a web portal (https://enesdataspace.vm.fedcloud.eu) including information, tutorials and training materials. It has recently been added to the Data Spaces radar, an initiative launched by the IDSA (International Data Space Association) with the main goal of mapping data spaces from several domains into one easy-to-use tool.The EDS is being employed in climate applications targeting big data processing, interactive analytics and visualization, and recently it has been extended to support more advanced scientific applications based on ML. In particular, it has been enhanced to provide a cloud-based development and testing environment for the implementation of some data-driven Digital Twin applications for extreme climate events in the context of the interTwin project. Finally, the ENES Data Space will be also one of the pilots in EOSC Beyond, a Horizon Europe project where the EDS will integrate and validate the new EOSC Core capabilities developed by the project.This work has been supported in part by interTwin; interTwin is funded by the European Union (Horizon Europe) under grant agreement No 101058386.

The advances in Earth System Models (ESM), jointly with the availability of more powerful computing infrastructures and novel solutions for Big Data and Machine Learning (ML) is allowing to push research in the climate change field forward. In such context, workflows are fundamental tools to automate the complex processes of model simulations, data preparation and analyses. Such tools are becoming more important as the complexity and heterogeneity in the software and computing infrastructures, as well as the data volumes to be handled, grow. However, integrating into a single workflow simulation and data centric processes can be very challenging due to their different requirements.This work presents an end-to-end workflow including the steps from the numerical ESM simulation run to the analysis of extreme weather events (e.g., heat waves and tropical cyclones) developed in the context of the eFlows4HPC EuroHPC project. It represents a real case study which requires components from High Performance Computing (HPC), Big Data and ML to carry out the workflow. In particular, the contribution demonstrates how the eFlows4HPC software stack can simplify the development, deployment, orchestration and execution of complex end-to-end workflows for climate science, as well as improve their portability over different computing infrastructures.

Oil spill incidents can have a significative negative impact on coastal and marine ecosystems as well as human activities. Despite increased maritime safety, the European Maritime Safety Agency's Cleanseanet program detected a substantial increase in spills within the Mediterranean basin. The accurate prediction of the transport and transformation of the oil slick is a key aspect for assessing the impacts of the spill on coastal and marine areas. In this context, numerical oil spill modeling plays a crucial role in understanding unseen impacts and filling observational gaps. Such models are executed according to a set of physical simulation parameters, which are usually hand-picked, relying mostly on the modeler's expertise. Proper selection of such parameters is key for ensuring accurate results. This study proposes a novel technique integrating satellite observations, the Medslik-II oil spill model and Machine Learning for enhancing the oil spill results by optimizing the model parametrization. A Bayesian Optimization Framework was implemented to search for the optimal configuration through the parameter space. A real oil spill case, that occurred in the Baniyas area (Syria) in 2021, was used to validate the proposed approach. Results from early evaluation of such framework are promising and demonstrated that coupling physics- and data-driven techniques can lead to more precise risk assessment and planning for oil spill incidents.

Current scientific workflow systems do not typically integrate simulation-centric and data-centric aspects due to their very different software/infrastructure requirements. A transparent integration of such components into a single end-to-end workflow would lead to a more efficient and automated way for generating insights from large simulation data. This work presents a complex case study related to extreme events analysis of future climate data that integrates in the same workflow numerical simulations, Big Data analytics and Machine Learning models. The case study is being implemented in the context of the eFlows4HPC project using the project’s software stack for deployment and orchestration of the workflow. The solution implemented in the project has shown to simplify the development and execution of end-to-end climate workflows with heterogeneous software requirements. Moreover, such an approach can, in the long term, increase the reuse of workflows by scientists and their portability over different HPC infrastructures.

Tropical Cyclones (TCs) are counted among the most destructive phenomena that can be found in nature. Every year, globally an average of 90 TCs occur over tropical waters, and global warming is making them stronger and more destructive. The accurate localization and tracking of such phenomena have become a relevant and interesting area of research in weather and climate science. Traditionally, TCs have been identified in large climate data sets through the use of deterministic tracking schemes that rely on subjective thresholds. This study presents a Machine Learning (ML) ensemble approach for locating TCs center coordinates. The ensemble combines TCs center estimates of different ML models that agree about the presence of a TC in input data. ERA5 reanalysis data was used for model training and testing jointly with the International Best Track Archive for Climate Stewardship (IBTrACS) records. Compared to single models estimates, the ML ensemble approach was able to improve TCs localization in terms of Euclidean Distance with respect to the observed TCs locations from IBTrACS. Moreover, a hybrid tracking scheme was defined: starting from the individual TC center locations detected by the ML ensemble approach, a deterministic tracking algorithm was used for reconstructing TC trajectories. The hybrid tracking scheme was then compared with four deterministic trackers reported in literature, achieving a Probability of Detection and a False Alarm Rate of 71.49% and 23%, respectively, over 40 years of reanalysis data.