Answers are grounded in this paper and cite their sources.
Coastal areas are increasingly vulnerable to rising sea levels, shoreline erosion, and more frequent extreme weather events. These evolving threats demand precise and timely vulnerability assessments to support effective planning and mitigation strategies. Yet, many existing prediction frameworks fall short-they often lack scalability, perform inconsistently across different coastal zones, and rely heavily on centralized data processing, which introduces delays and hinders real-time responsiveness. To overcome these limitations, this study proposes FGMT-Net, a novel Federated Graph-Masked Transformer Network designed specifically for dynamic, zone-sensitive coastal vulnerability prediction. Rather than focusing solely on data privacy, FGMT-Net uses federated learning to enable low-latency, decentralized training across regional data sources-such as Aberdeen Bay, Stonehaven, and St. Andrews-without requiring data centralization. The model architecture blends three core innovations: (1) graph-based spatial learning to capture cross-zone dependencies, (2) temporal transformers to model evolving environmental patterns, and (3) an attention-guided feature masking mechanism that enhances both interpretability and noise reduction. Our primary objective is to build a scalable, interpretable, and communication-efficient framework for multi-target coastal risk forecasting. Experimental results show that FGMT-Net delivers strong performance, achieving a 97.6% classification rate, converging within 18 communication rounds, and maintaining high consistency across regions-reflected in FAQ and PAC scores of 0.931 and 0.974, respectively. Additionally, statistical evaluations (ANOVA F = 8.90, Pearson r = 0.95, Spearman <tex-math notation="LaTeX">ρ</tex-math> = 0.93) confirm its robustness. The novelty of this work lies in fusing federated deployment with graph-masked transformer reasoning, paving the way for fast, transparent, and zone-aware coastal vulnerability monitoring.
No takes yet. Share an insight, caveat, or question.
Qi et al. (2025) studied this question.
Synapse has enriched 10 related papers. Consider them for comparative context: