Thematic Insights into Climate Change and Cultural Heritage via Topic Modeling and Semantic Embeddings

Climate change poses escalating threats to cultural heritage worldwide, yet scholarly research on this intersection remains fragmented across disciplines and methodologies. This study presents the first large-scale, reproducible thematic mapping of the climate–heritage literature, synthesizing 850 peer-reviewed articles published between 2016 and 2024 through a PRISMA-guided systematic review. We integrated six topic modeling techniques—Correlated Topic Model, Hierarchical Dirichlet Process, Latent Dirichlet Allocation, Latent Semantic Indexing, Non-negative Matrix Factorization, and Structural Topic Model—with BERT-based semantic embeddings and supervised classifiers (i.e., Random Forest and ensemble methods) to derive a 32-topic taxonomy, consolidated into eight overarching themes. The combination of NMF-derived topic probabilities and contextual embeddings achieved classification accuracies up to 98%, with minimal performance loss on a temporally independent validation set of 259 articles. Key themes include flood-related hazards, air pollution impacts, risk and vulnerability assessments, indoor environmental quality in museums, and adaptation strategies. Overlaps between semantically adjacent categories reflect genuine conceptual convergence in the literature. Beyond methodological innovation, this framework offers actionable insights for policy and conservation. It highlights under-represented domains—such as heritage–climate justice. Future work will extend analysis to full-text corpora, incorporate multi-annotator labeling, adapt domain-specific language models, and implement dynamic ontologies for real-time monitoring. This integrative approach establishes a scalable foundation for evidence-based decision-making to safeguard cultural heritage under accelerating climate risks.