Systematic Literature Review of Graph Neural Networks in Disaster Management: Methods, Applications, and Future Directions
Sularno1*, Wendi Boy2, Putri Anggraini3, Rometdo Muzawi4, Renita Astri5
1Department of Information System, Faculty of Information System, Dharma Andalas University, Padang, Sumatera Barat 25000, Indonesia, Email: soelarno@unidha.ac.id
2Department of Civil Engineering, Faculty of Civil Engineering, Dharma Andalas University, Padang, Sumatera Barat 25000, Indonesia, Email: wendi@unidha.ac.id
3Department of Information System, Faculty of Information System, Dharma Andalas University, Padang, Sumatera Barat 25000, Indonesia, Email: bontetga@unidha.ac.id
4Department of Information Technology, Faculty of Information Technology, University of Indonesia Science and Technology, aridua, Meghalaya 793101, Indonesia, Email: rometdomuzawi@usti.ac.id
5Department of Information System, Faculty of Information System, Dharma Andalas University, Padang, Sumatera Barat 25000, Indonesia, Email: rethakamal@unidha.ac.id
Abstract
The increasing frequency and complexity of natural disasters have heightened the urgency for more intelligent, adaptive, and data-driven disaster management systems. This study presents a Systematic Literature Review (SLR) on the application of Graph Neural Networks (GNNs) in disaster management, aiming to provide a comprehensive synthesis of current methodologies, applications, and research gaps. Employing the PRISMA 2020 framework and PICOC formulation, a total of 4,087 studies from IEEE, Scopus, SpringerLink, and ACM Digital Library were screened, resulting in 50 relevant articles published between 2019 and 2025. The review reveals that GNNs, particularly Graph Convolutional Networks (GCNs), Graph Attention Networks (GATs), and hybrid models like GNN-GRU and GNN Transformer, have been widely implemented across various disaster phases including mitigation, emergency response, and post-disaster recovery. GNNs are primarily applied in flood and wildfire prediction, evacuation planning, infrastructure damage assessment, and pandemic modeling. Their ability to model spatiotemporal relationships makes them effective tools for handling complex disaster-related data. However, key challenges persist, including issues of scalability, data quality, model interpretability, and limited cross-disaster generalizability. To address these, future research should explore more interpretable and scalable architectures, improved integration with IoT systems, and the development of general-purpose GNN models for multi-disaster scenarios. This review contributes valuable insights for researchers, policymakers, and system developers aiming to build transparent, responsive, and robust AI-based disaster management frameworks.
Keywords: Disaster Management; GAT; GCN; Graph Neural Networks; Systematic Literature Review