Legacy scientific codes are essential to long-standing operational workflows but often cannot fully exploit modern heterogeneous high-performance computing (HPC) architectures. GLOBO, a global atmospheric circulation model developed at CNR-ISAC, is one such model that is still used for research and operational weather forecasting. In this work, we present an ongoing refactoring of GLOBO to improve its scalability and prepare it for GPU acceleration and cloud execution. The modernization focuses on three main areas: dynamic memory allocation, replacement of legacy point-to-point MPI communications with optimized collective operations, and OpenACC-based GPU offloading. We describe the redesign of the communication strategy, which replaces multiple MPI_Isend/MPI_Irecv exchanges with collective primitives (MPI_Bcast, MPI_Scatterv, MPI_Sendrecv, MPI_Reduce) to reduce overhead and improve parallel efficiency. Scalability experiments across multiple spatial resolutions show that a new collective communication strategy consistently matches or exceeds the original implementation, delivering up to 40% runtime reduction in high-resolution, multi-node configurations. These results lay the groundwork for refactoring GLOBO toward a fully dynamic, GPU-accelerated, cloud-ready execution, preserving its scientific heritage while ensuring sustainable performance on future HPC platforms.

Global changes are increasing the frequency and intensity of extreme weather events, posing challenges for forecasting localized phenomena with sub-grid resolution. The Hi-WeFAI project addresses this by combining high-performance computing, Federated Artificial Intelligence, and heterogeneous sensor networks to improve short-term precipitation forecasting and flood nowcasting. In this paper, we present preliminary results using a transformer-based radar prediction model coupled with a flood model to generate high-resolution early warning maps. The results from the Naples pilot site improved accuracy and detail, highlighting the potential of the hybrid AI and HPC approach to support quasi-real-time decision-making and disaster risk reduction.

This paper introduces DAGonCAPIO, a workflow framework that integrates the DAGonStar engine with the CAPIO middleware to enable I/O streaming in scientific workflows. DAGonStar utilizes a Directed Acyclic Graph (DAG) to orchestrate tasks. At the same time, CAPIO enables downstream tasks to process data as soon as partial outputs become available, without requiring modifications to application code. This integration reduces delays from traditional file-based communication. The system uses the workflow:// schema to define data dependencies and generate CAPIO coordination scripts. DAGonStar was modified to support early scratch directory naming and decoupled task execution. Experiments with a WRF-based weather forecasting workflow on a 256-core cluster demonstrate that DAGonCAPIO reduces time-to-first-result by up to 4171 seconds, achieving a nearly 10x speedup.

Marine litter detection is crucial for environmental monitoring, yet the imbalance in existing datasets limits model performance in identifying various types of waste accurately. This paper presents an efficient data augmentation pipeline that combines generative diffusion models (e.g., Stable Diffusion) and Large Language Models (LLMs) to expand the G-Litter dataset, a marine litter dataset designed for autonomous detection in heterogeneous environments. Leveraging scalable diffusion models for image generation and Alpaca LLMs for diverse prompt generation, our approach augments underrepresented classes by generating over 200 additional images per class, significantly improving the dataset’s balance. Training G-Litter augmented dataset using YOLOv8 for object detection demonstrated an increase in detection performance, improving recall by 7.82% and mAP50 by 3.87% (compared with baseline results). This study emphasizes the potential for combining generative AI with HPC resources to automate data augmentation on large-scale, unstructured datasets, particularly in edge computing contexts for real-time marine monitoring. The models were tested on real videos captured during simulated missions, demonstrating a superior ability to detect submerged objects in dynamic scenarios. These results highlight the potential of generative AI techniques to improve dataset quality and detection model performance, laying the foundation for further expansion in real-time marine monitoring.

Abstract The request for quickly available forecasts of intense weather and marine events impacting coastal areas is gradually increasing. High-performance computing (HPC) and artificial intelligence techniques are crucial in this application. Risk mitigation and coastal management must design scientific workflow appropriately and maintain them continuously updated and operational. Climate change accelerating increase trend of the past decades impacted on sea-level rise, together with broader factors such as geostatic effects and subsidence, reducing the effectiveness of coastal defenses. Due to this, the support tools, such as Early Warning Systems, have become increasingly more valuable because they can process data promptly and provide valuable indications for mitigation proposals. We developed the Shoreline Alert Model (SAM), an operational Python tool that produces simulation scenarios, ‘what-if’ assumptions, and coastal flooding forecasts to fill this gap in our study area. SAM aims to provide decision-makers, scientists, and engineers with new tools to help forecast significant weather-marine events and support related management or emergency responses. SAM aims to fill the gap between the wind-driven wave models, which produce simulations and forecasts of waves of significant height, period, and direction in deep or mid-water, and the run-up local models, which exstimulate marine ingression in the event of intense weather phenomena. It employs a parallelization scheme that allows users to run it on heterogeneous parallel architectures. It produced results approximately 24 times faster than the baseline when using shared memory with distributed memory, processing roughly 20,000 coastal cross-shore profiles along the coastline of the Campania region (Italy). Increasing the performance of this model and, at the same time, honoring the need for relatively modest HPC resources will enable the local manager and policymakers to enforce fast and effective responses to intense weather phenomena.

Autonomous Surface Vehicles and their management represent one of the significant challenges in coastal and offshore surveying. Although the development of this kind of data acquisition device has skyrocketed in the last few years, line guides and technological solutions still need to come. On the other hand, this kind of robotic vessel's true potential has yet to be explored. This paper presents ArgonautAI, a containerized distributed processing platform for autonomous surface vehicles. The proposed ArgonautAI architecture leverage a cluster of single-board computers with diverse and different characteristics (computing power, CUDA GPUs, FPGAs, GPIOs, PWMs, specialized I/O) orchestrated using Kubernetes and a customized programming interface. Furthermore, the proposed solution introduces two different types of containers: 1) the platform containers hosting the software life support for the platform and 2) the mission containers defined to support the survey mission-specific scopes. The firsts manage the vehicle's instruments (e.g. position, attitude, environment, depth), the data storage, the vessel-to-shore communication, and so on; the latter host mission-specific software components. Finally, as proof of concept of the proposed platform, we present an AI-based marine litter detection application using a hierarchical computer vision approach on heterogenic onboard computing resources.

While using High-Performance Computing (HPC) for precise and accurate air quality forecasts is a common issue, similar services devoted to marine pollution in coastal areas remain challenging. This paper presents Water quality Community Model Plus Plus (WaComM++) leveraging a parallelization schema enabling the users to run it on heterogeneous parallel architectures. We evaluated the proposed model under several execution approaches using a real-world application for pollutants forecast in the Gulf of Napoli (Campania, Italy). As a result, WaComM++ has produced results 657K times faster than the sequential run (taking into account the Particles' Outer Cycle and not considering the particle domain distribution) when using distributed and shared memory with multi-GPUs dealing with about 25 million particles.

In this paper, the Shoreline Alert Model (SAM) is presented as a component of a computation platform based on workflows dedicated to extreme weather/marine event simulation. The model aims to mitigate the effects of global change by providing decision-makers, scientists, and engineers with a novel, next-generation tool set for facing extreme weather events and implementing related management or emergency responses. SAM uses a parallelization schema, allowing users to run it on heterogeneous parallel architectures. As a result, SAM produces approximately 24 times faster results than the baseline when using shared memory with distributed memory and dealing with about 20,000 transects along the Campania coastline. The system is based on the algorithms of the open-source numerical models WRF (Weather Research and Forecasting) and WW3 (Wave-watch III) implemented with refraction and shoaling routines together with run-up equations to form the modeling chain used for coastal flooding assessment.

The use of hardware accelerators, based on code and data offloading devoted to overcoming the CPU limitations in cores, is one of the main distinctive trends in high-end computing and related applications in the last decade. However, while code offloading is convenient for performance improvement, becoming a commonly used paradigm, memory access and management are a source of bottlenecks due to the need to interact with different address spaces. In this regard, NVidia introduced the CUDA Unified Memory model to avoid explicit memory copies between the machine hosting the accelerator device and the device itself and vice-versa. This paper shows a novel design and implementation of the support to the CUDA Unified Memory in open-source GPGPU virtualization services. The performance evaluation demonstrates that the overhead due to the virtualization and remoting is acceptable considering the possibility of sharing CUDA-enabled GPUs between various and heterogeneous machines hosted at the edge, in cloud infrastructures, or as accelerator nodes in an HPC scenario. A prototype implementation of the proposed solution is available as open-source.

Monitoring the impact of the pollutants on the sea is a crucial issue for coastal human activities, such as aquaculture. However, leveraging a continuous microbiological laboratory analysis is unfeasible for costs and practical reasons. Here we present a novel methodology finalized to predict water quality as categorized indexes leveraging an integrated approach between computational components and artificial intelligence techniques. As a paradigm demonstrator, we couple WaComM++ with AIQUAM. The use case presented is an application of AIQUAM in the Bay of Naples (Campania Region, Italy) for predicting bacteria contaminants in mussel farms. The results are encouraging as the model reached a correct prediction rate of 93%.

In coastal areas, the wind vector at sea is one of the most important geophysical variable of interest. Wind field information is crucial for navigation, green energy production, fishing, etc. Nonetheless, to obtain distributed and accurate sea wind information in coastal areas is a challenging task that cannot be achieved by buoys, anemometers or satellite sensors as lidars and scatterometers. Hence, in this study, the behavior of sea wind field is analyzed over the selected test site, i. e., the Gulf of Naples (Italy, Mediterranean Sea) using level-2 ocean products obtained by the C-band Synthetic Aperture Radar (SAR) on board of the European Space Agency mission Sentinel-1. The latter are compared with space/time co-located wind vectors obtained by the Weather and Research Forecasting (WRF) model. Preliminary experimental results indicate that there is a fairly good agreement between wind vectors estimated from SAR and the ones predicted by the WRF model. They suggest that a moderate eastern wind blowed over the study area in the considered period.

Summary The rise of the Internet of Things has generated high expectations about the improvement in people's lifestyles. In the last decade, we saw several examples of instrumented cities where different types of data were gathered, processed, and made available to inspire the next generation of scientists and engineers. In this framework, sensors and actuators became leading actors of technologically pervasive urban environments. However, in coastal areas, marine data crowdsourcing is difficult to apply due to the challenging operational conditions, extremely unstable network connectivity, and security issues in data movement. To fill this gap, we present a novel version of our DYNAMO transfer protocol (DTP), a platform‐independent data mover framework where data collected on board of vessels are stored locally and then moved from the edge to the cloud when the operating conditions are favorable. We evaluate the performance of DTP in a controlled environment with a private cloud by measuring the time it takes for the clouds ide to process and store a fixed amount of data while varying the number of microservice instances. We show that the time decreases exponentially when the number of microservice instances goes from 1 to 16 and it remains constant above that number.

Summary Fast technology development has influenced the widespread use of low‐power devices in different scientific, environmental, and everyday life areas, giving birth to the Internet of Things. In this paper, we focus on the context of marine studies, addressing the problem of marine bathymetry data processing and analysis via pervasive and Internet‐connected sensors and low‐power distributed devices. Pervasive and Internet‐connected low‐power devices (as the components involved in the sensing and processing actions) made diverse and different “things” as a worldwide‐distributed system. Given the high complexity of the algorithms involved in these studies, which usually involve general‐purpose graphic processing unit (GPGPU) computation, it is impossible for the limited devices to perform the required calculations. To overcome these limitations, in this paper, we propose and implement a vertical application of GVirtuS, the open‐source GPGPU virtualization and remoting service, for achieving high performance geographical data interpolation in a high performance cloud computing scenario. We present an innovative implementation by comparing, in terms of performance and accuracy, the inverse distance weighting and kriging interpolation methods in their parallel implementations leveraging on CUDA‐enabled GPGPUs. We present a real‐world use case related to high‐resolution bathymetry interpolation in a crowdsource data context in the Bay of Pozzuoli, Italy.

Summary Low‐power devices are usually highly constrained in terms of CPU computing power, memory, and GPGPU resources for real‐time applications to run. In this paper, we describe RAPID, a complete framework suite for computation offloading to help low‐powered devices overcome these limitations. RAPID supports CPU and GPGPU computation offloading on Linux and Android devices. Moreover, the framework implements lightweight secure data transmission of the offloading operations. We present the architecture of the framework, showing the integration of the CPU and GPGPU offloading modules. We show by extensive experiments that the overhead introduced by the security layer is negligible. We present the first benchmark results showing that Java/Android GPGPU code offloading is possible. Finally, we show the adoption of the GPGPU offloading into BioSurveillance, a commercial real‐time face recognition application. The results show that, thanks to RAPID, BioSurveillance is being successfully adapted to run on low‐power devices. The proposed framework is highly modular and exposes a rich application programming interface to developers, making it highly versatile while hiding the complexity of the underlying networking layer.