http://dspace.azti.es:8080 The DSpace digital repository system captures, stores, indexes, preserves, and distributes digital research material. 2026-07-25T00:50:59Z http://dspace.azti.es/handle/24689/2794 Title: Towards Trustworthy Artificial Intelligence for Marine Research, Fisheries and Environmental Management Authors: Fernandes-Salvador, Jose A; Borja, Angel; Anabitarte, Asier; Granado, Igor; Lekunberri, Xabier; Sagarminaga, Yolanda; Canals, Oriol; Lanzen, Anders; Azhar, Mihailo; Kotta, Jonnen; Ojaveer, Henn; Spinosa, Anna; Jokinen, Ari-Pekka; Haraguchi, Lumi; Stæhr, Sanjina Upadhyay; Pérez, Aritz; Inza, Iñaki; Villasante, Sebastian; Oanta, Gabriela A.; Silva, Catarina N. S.; Tiller, Rachel; Lilkendey, Julia Abstract: Artificial Intelligence (AI) is advancing at an unprecedented pace, offering transformative opportunities for marine research, fisheries management, environmental governance and policy development. Particularly in the context of the interconnected data needs of ecosystem management and biodiversity conservation, these technologies can enhance data acquisition, processing and decision support, enabling more integrated approaches to ecosystem management and biodiversity conservation. Yet their adoption in these domains remains limited by the absence of coherent frameworks that ensure transparency, validation and ethical alignment with ecological and socio-economic sustainability goals. This work proposes a comprehensive framework built on three critical pillars for trustworthy AI: socio-economic and legal viability, data governance and technical and scientific robustness. On the one hand it aims to be a guideline for developer teams. On the other hand, it aims to be a guideline for final users (e.g., industry and managers) for designing the requirements and evaluating such systems. The first pillar underscores the need for AI systems that are cost-effective, scalable, environmentally sustainable and legally supported, balancing short-term costs with long-term social and ecological benefits. The second stresses adherence to fair, reliable and ethical access to digital resources, recognising that without strong governance data and algorithms risk becoming fragmented or misused. The third pillar addresses the necessity of rigorous validation across entire AI pipelines, including preprocessing, model evaluation and benchmarking against alternative ground truths, to ensure reliability in real-world applications. Together, these pillars provide a blueprint for developing ethical, reliable and policy-relevant AI systems that can strengthen trust, improve sustainability and guide decision-making across marine science, fisheries, environmental management and European legislation. 2026-07-24T00:00:00Z http://dspace.azti.es/handle/24689/2790 Title: HFR-Granitola System Reports Authors: Solabarrieta, Lohitzune Lohitzune 2026-11-01T00:00:00Z http://dspace.azti.es/handle/24689/2789 Title: HFR-GoS System Reports Authors: Solabarrieta, Lohitzune 2026-11-01T00:00:00Z http://dspace.azti.es/handle/24689/2785 Title: Evolutionary multi-objective fishing routing with decision maker's preferences Authors: Granado, Igor; Szlapczynska, Joanna; Szlapczynski, Rafal; Hernando, Leticia; Fernandes-Salvador, Jose A. Abstract: This study aims to enhance economic and environmental sustainability of fisheries through fishing routing methods that can reduce operational costs, emission footprints, and incidental fishing risks. To achieve this, a novel problem definition is introduced, the time-dependent multi-objective orienteering problem with time windows and moving targets (TDMOOP-TWMT). Unlike existing fishing routing problems, the TDMOOP-TWMT allows users to define their fishing trips by setting a maximum time at sea rather than a predefined number of fishing sets. This multi-objective problem includes three goals: fuel-oil consumption, catches of tuna species, and incidental catches of non-target species (bycatch). To address this problem, the w-MOEA/D algorithm is employed, which incorporates decision-makers' preferences using wide weight intervals for each objective, eliminating the need for precise weight values. Compared to the classical MOEA/D, the w-MOEA/D method achieves solutions closer to the true Pareto front while reducing the final solution set based on users' preferences. To demonstrate the potential application and benefits in a real context, 12 historical routes are employed across different fishing scenarios, each defined by varying the weight intervals of the objectives. The results show that w-MOEA/D routes allow for consuming less fuel and catching more tuna, though with a higher risk of bycatch when compared to historical trips. However, prioritizing bycatch avoidance reduces this risk while maintaining similar fuel efficiency, although with a lower increase in catches. In summary, this study highlights the effectiveness of the proposed solution method in supporting fishers' decision-making by incorporating their preferences when planning fishing routes. 2025-06-30T00:00:00Z