The increasing impact of global change on coastal areas necessitates effective collaboration among stakeholders to develop sound environmental protection policies. However, the vast number of scientific publications and the interdisciplinary nature of the field make it challenging to accurately identify and extract relevant information. Current methods for analyzing coastal region literature are limited in handling complex document-level relations, impeding the development of comprehensive Knowledge Bases (KBs) essential for informed decision-making. Therefore, we introduce CoastRED, a corpus of 416 scientific abstracts focused on coastal regions, designed for interdisciplinary end-to-end Document-level Relation Extraction (DocRE), leveraging the ARDI framework (Actors, Resources, Dynamics, and Interactions) to automatically identify key entities and their roles in coastal systems. We present DUNJERE—an end-to-end model integrating Mention Detection, Coreference Resolution, Entity Typing, and Relation Extraction. Based on a U-Net architecture, DUNJERE approaches Relation Extraction as a semantic segmentation task, enhancing its performance in an end-to-end setting. Evaluations on the DocRED and ReDocRED datasets demonstrate state-of-the-art performance, while experiments on CoastRED highlight its potential to advance coastal system research by providing a robust tool for document analysis and KB construction. Code and dataset are available on this repository: https://github.com/jdelaunay/coastred .