This work presents a method for generating and refining a knowledge graph (KG) from a historically significant 20th-century marine biology text. The source, a foundational ecological survey, was digitized using OCR and processed for semantic consistency. Knowledge extraction was performed with GraphRAG and the GPT-4o-mini model, producing an initial KG with verbose and inconsistent relationships. To improve clarity and alignment with semantic web standards, a two-step refinement process was applied, combining automated tuning and prompt engineering. The result was a set of concise, RDF-style predicates suitable for querying and integration with ontological frameworks. The refined KG is accessible via a public platform supporting multilingual natural language queries, enabling broader use of historical ecological data. This approach highlights the potential of AI-assisted pipelines to transform legacy scientific texts into semantically structured, interoperable resources for biodiversity research.