How deep learning shapes disaster research: a bibliometric analysis

The use of deep learning (DL) in disaster research has rapidly emerged as a key approach. In this study, we systematically compiled 247 scientific publications using the terms DL and disaster, published between 2016 and 2025, using the Scopus database. The analysis, conducted using the R-based Bibliometrix software, focused on multidimensional indicators such as annual publication trends, co-authorship networks, citation patterns, conceptual clusters, and the temporal evolution of task types. The findings indicate that disaster-themed DL research is concentrated around specific countries and institutions, with publication clusters diversifying over time and evolving into more methodologically integrated structures. In particular, we observed that early concepts such as detection and classification have evolved into more complex concepts such as segmentation, assessment, and spatial mapping in recent years. Furthermore, significant thematic relationships were identified among task clusters, disaster types, and regional clusters. The findings also indicate that DL has evolved into a key technical tool in disaster management and an increasingly influential methodological approach shaping patterns of scientific production.