A recommender framework for sustainable coastal disaster management using big data, maritime surveillance, and AIOps

The convergence of intensifying human activity and climate change is escalating the frequency, magnitude, and complexity of maritime disasters, while real-time situational awareness and coordinated response remain critical bottlenecks. This paper presents a recommender framework and its underlying Data and AI Integration Platform for sustainable coastal disaster management, integrating maritime surveillance, big-data pipelines, and MLOps/AIOps automation. The core technical contribution is a dual-path decision-support engine that couples a lightweight 1D-CNN time-series event classifier with a fine-tuned BART large language model, enabling simultaneous reasoning over structured sensor telemetry and unstructured maritime text. The classifier is validated on real-world AIS telemetry (MarineCadastre.gov) aligned with NOAA buoy data as well as a controlled synthetic corpus, attaining 93.9% accuracy and a 92.4% macro-F1 at ~ 5.3 ms per sample, and is benchmarked against LSTM, Bi-LSTM + Attention, Temporal Convolutional Network (TCN), Transformer, and Temporal Fusion Transformer (TFT) baselines. The BART summariser, fine-tuned on 2000 maritime documents, reaches ROUGE-1 73.4, ROUGE-L 68.7, and BLEU 65.2, outperforming GPT-2 and T5 baselines, and is assessed by five domain experts (Fleiss’ kappa = 0.71). Beyond component metrics, we report platform-level indicators, including end-to-end alert latency, situational-awareness gain, and analyst reading-time reduction, to demonstrate the operational benefit of the integrated system for Coast Guard and maritime authorities. The platform-level indicators are obtained from an explicitly documented replay protocol over 320 annotated incident scenarios of stated provenance, the real-world AIS labels are produced by a two-stage rule-plus-expert annotation procedure (Fleiss’ kappa = 0.78), and every model comparison is reported with bootstrap confidence intervals and McNemar significance tests. Results indicate that the framework advances the shift from reactive to proactive, transparent, and resilient maritime disaster response.