Trustworthy urban digital twin: a RAG-based architecture for integrating verifiable knowledge

Urban Digital Twins (UDTs) can revolutionise city management, yet their adoption is constrained by a trust deficit, largely due to the lack of contextual transparency in purely data-driven outputs. This paper argues that achieving trustworthy UDTs requires a rethinking of the architectural shift from opaque data pipelines to a framework grounded in verifiable knowledge. To address this, we propose a UDT architecture that leverages Large Language Models (LLMs) and Retrieval-Augmented Generation (RAG). In this architecture, RAG acts as a "trust anchor", retrieving immutable, source-linked evidence from a curated knowledge graph, while the LLM serves as a "transparent synthesiser", reasoning exclusively upon this retrieved evidence with an auditable thought process. The proposed solution is demonstrated through a case study within the Newcastle Clean Air Zone (CAZ). A knowledge graph, constructed from local policy and scientific reports, is used to interpret and contextualise air quality data from the Newcastle Urban Observatory. This testing successfully demonstrates how the proposed architecture addresses complex urban inquiries in a manner that is auditable, explainable, and grounded in verifiable facts. This paper establishes a blueprint for transforming UDTs into transparent, evidence-grounded systems that foster trust among planners, policymakers, and the public.