Beyond Physical Observation: AI Based Crisis Event Sensing from Heterogeneous News Streams for Environmental Situational Awareness

Introduction: Remote sensing and Earth observation provide essential evidence for monitoring disasters, environmental hazards, infrastructure disruptions, and emergency events. However, crisis signals also emerge through heterogeneous digital news streams, which can complement physical observation by providing rapid contextual information. This study develops an AI-based crisis-event-sensing framework that treats online news as a complementary digital sensing layer for environmental situational awareness.

Methods: The study analyzes 71,930 valid crisis news stories, covering 20 event categories and 772 source domains, collected over a continuous 32-month period between 26 September 2023 and 02 June 2026 via automated RSS feeds, News API outputs, and web scraping to capture multi-year temporal variations. The framework integrates acquisition exposure normalization, source-event bipartite network modeling, event similarity analysis, imbalanced multiclass classification, feature ablation, and rare-class transfer diagnostics. Temporal interpretation is corrected using active source domain days, while source-event networks quantify reporting concentration and relational proximity among disaster, infrastructure, transport, industrial, and emergency categories.

Results: The results show that raw crisis news volume can misrepresent environmental monitoring trends. Mean monthly raw records declined by 79.26% from the baseline to the recent period, but active source domains declined by 89.62% and active source domain days by 76.90%. After exposure normalization, the decline in reporting intensity was only 10.64%. The source-event network contained 20 event nodes, 772 source nodes, and 2,540 weighted edges, indicating a sparse but informative sensing structure. Classification experiments on 25,000 stratified records showed that the proposed full representation model achieved the best macro F1 of 0.564 and balanced accuracy of 0.559, outperforming classic baseline models including BERT and CNN. Rare-class transfer improved Significant Earthquake Event from 0.00 to 0.47 F1 and National Infrastructure Event from 0.00 to 0.19 F1.

Conclusions: This study demonstrates that heterogeneous news streams complement remote sensing as an AI-enabled digital event-sensing layer. The framework supports environmental situational awareness, disaster monitoring, emergency intelligence, and acquisition-corrected crisis analytics without equating raw media volume with real-world event prevalence.