The Evolution of Spatiotemporal Hotspot Prediction in Public Health: Emerging Paradigms from Spatial Epidemiology to GeoAI and Foundation Models

Abstract

Spatiotemporal hotspot prediction has become essential for proactive public health surveillance and intervention. Over several decades, analytical approaches in this field have evolved from classical spatial epidemiology through Bayesian modeling, machine learning, deep learning, GeoAI, and more recently, foundation models. This State-of-the-Art Review systematically traces the evolution of this paradigm, comparing methodological characteristics, strengths, and limitations across successive approaches. This State-of-the-Art Review synthesizes evidence from peer-reviewed studies identified through a structured literature search. It identifies key patterns of advancement, including increasing predictive capability, the re-integration of spatial reasoning, and a growing emphasis on operational scalability and multimodal intelligence. It also highlights persistent challenges related to interpretability, data equity, computational demands, and governance. The synthesis reveals that the current state of the art is defined by the convergence of GeoAI and foundation models, which together offer new possibilities for context-aware and generalizable public health intelligence. The review concludes by outlining evidence-supported future trajectories and emphasizing that continued progress will depend on addressing both technical and institutional challenges.