Synthesising textual- and model-based evidence for proactive urban policymaking: A case in air quality management

Urban governance is hindered by the disconnect between qualitative knowledge embedded in textual sources and the predictive capabilities of quantitative models, making proactive policymaking challenging. This study proposes and validates a two-phase framework that automates the 'dialogue' between the evidence paradigms for urban air quality management. In the first phase, a Retrieval-Augmented Generation (RAG) pipeline synthesises extensive literature into a structured knowledge graph, generating evidence-based policy hypotheses. This approach navigates the challenge of evaluating overwhelming potential policy combinations. Following this, these qualitative hypotheses are automatically translated into simulation scenarios and subjected to rigorous, ex-ante quantitative testing, providing the rapid feedback necessary to overcome slow policy evaluation cycles. A case study in London, addressing post-2023 ULEZ expansion air quality challenges, demonstrates the framework's real-world utility. The RAG identified a portfolio of next-generation interventions targeting heating systems, construction machinery, and solvent emissions. Subsequent simulation with the SHERPA model produced spatial impact maps of pollution-level reductions and, critically, uncovered complex system dynamics. The framework's ability to efficiently screen a broad portfolio of options while revealing such complex consequences underscores its value for proactive governance. By integrating textual and simulation evidence into a scalable methodology, this research provides a robust tool to enable more effective, evidence-informed urban environmental policy.