AI-Enabled Geospatial Disaster Intelligence from Global News Streams: A Soft Sensing Framework for Natural Hazard Risk Analytics

Introduction:
Rapidly evolving natural hazards require Earth science systems capable of detecting risk signals beyond conventional satellite, meteorological, and ground-based observations. This study introduces an AI-enabled geospatial disaster intelligence framework that treats global news ecosystems as a scalable soft sensing layer for extracting, structuring, and analysing geographically distributed disaster risk signals. The framework advances Earth science analytics by integrating large-scale artificial intelligence with spatial hazard intelligence for improved situational awareness and risk interpretation.

Methods:
The study analyses a large disaster intelligence corpus generated from 1.25 million global news articles collected over 514 days from 444 international news portals. Using GPT-based semantic extraction, 17,884 disaster-related reports were autonomously identified and transformed into structured geospatial event records. The resulting dataset captured disaster type, affected country, location, temporal markers, and severity indicators, covering 185 countries and 6,068 unique locations. Topic modelling and statistical inference were then applied to quantify how hazard narratives vary across disaster categories and geographical contexts.

Results:
The framework revealed robust and interpretable geospatial patterns of disaster risk. Statistical testing showed that narrative structure differed significantly by disaster type, χ² = 25,280.78, p < 0.001, and by affected country, χ² = 23,564.62, p < 0.001. Distinct hazard signatures emerged across regions, including earthquake-centred intelligence in Japan and Myanmar, hurricane-linked signals in the United States, and wildfire and flood-related risk patterns across multiple climate-sensitive areas. These findings demonstrate that AI extracted news signals can uncover statistically significant spatial heterogeneity in disaster risk communication and event concentration.

Conclusions:
The study establishes a novel GeoAI-based soft sensing paradigm for Earth science and natural hazard analytics. By converting global news streams into structured geospatial intelligence, the proposed framework complements physical Earth observation systems and supports more responsive disaster monitoring, cross-regional risk assessment, and early warning-oriented decision support. The approach offers a scalable pathway for integrating AI, big data, and geographic intelligence into next-generation Earth science applications.