Web-Based Severity Assessment of Natural Hazards: Natural Language Processing Based Extraction of Severity Impact Factors for Informed Decision Support

Abstract

This study introduces an automated information extraction (IE) method for assessing natural hazard severity using online sources (news articles and social media). A web crawler collects 4-15 daily news articles from diverse web sources, amassing 4,000 reports on natural hazard events between 05/2020 and 11/2023, while adhering to respective web page privacy rules. The extracted data analyses hazard severity, focusing on the impact factors of casualties and damages. This analysis aids decision-makers and researchers in comprehending the impact of hazards and developing mitigation strategies. Real-time web data severity analysis can also support first responders in resource allocation during and post-disasters. A natural language processing-based algorithm identifies hazard impact factors using grammatical patterns, PoS (Parts of Speech) tagging, and NER (Named Entity Recognition). From these, we identify numeric values for casualties, infrastructural and financial damages and public necessities, along with place names. The data is structured in a database for analysis. We utilize TOPSIS (Technique for Order of Preference by Similarity to Ideal Solution), a Multi-Criteria Decision-Making method, to assign a severity rank to each location. In TOPSIS, we define the positive ideal solution as the maximum values for positive attributes (e.g., number of rescue operations, people rescued, and government authorities’ involvement). The negative ideal solution represents the minimum values for negative attributes (e.g., damages and fatalities) in each criterion. We then calculate relative closeness (0.0 to 1.0) by measuring each criterion’s distance from the positive and negative ideal solutions. A higher relative closeness indicates less severity, while a lower value suggests greater severity in hazard events. This rank aids in identifying the area with the highest severity, enabling first responders to allocate resources effectively by prioritizing the locations with the most significant impact. Each location’s severity rank is based on relative closeness to positive and negative ideal solutions. We apply our methodology to the 2018 Kerala, India floods, using 200 news reports (national and local news portals, blogs), identifying the Alappuzha district as the most severely affected (highest severity score) and the Kasargod district the least affected (lowest severity score). Agricultural loss emerges as a significant factor, emphasizing the need for sustainable solutions. Results are consistent with official Kerala State Disaster Management Authority documentation, demonstrating the methodology’s accuracy. Our methodology provides near real-time information for identifying and prioritizing severely affected areas, aiding efficient resource allocation, rehabilitation efforts, and post-disaster decision-making.