Eine Plattform für die Wissenschaft: Bauingenieurwesen, Architektur und Urbanistik
Eine Plattform für die Wissenschaft: Bauingenieurwesen, Architektur und Urbanistik
Each year, tropical cyclones (TCs) cause significant damage to both society and the environment through multiple hazards. While extensive research has focused on individual TC primary hazards, there is a gap in the comprehensive assessment of multiple TC-related hazards using a large number of TCs, likely due to limitations in methodology and data when identifying complex hazard interrelationships. To address this challenge, we developed novel methods using a rule-based natural language processing (NLP) approach to extract TC-related weather-hazardous events from official hazard records, of which narratives provide detailed characterization of TC multi-hazard interrelationships. We applied this method on TCs originating from the North Atlantic and East Pacific Oceans affecting the Contiguous United States. The results showed that the NLP methods, especially the large language models, can identify TC names from the texts with reasonable precision. A total of 21 488 events related to 179 TCs were identified from 2007 to 2022. These events encompassed 31 types of hazards, with precipitation and flooding being the most frequent and causing the most direct fatalities, while wind hazards cause the most direct property damage. Although TC wind and storm surges are typically concentrated within 100–200 km of the storm track, hazards can occur hundreds of kilometers away and even after the storm’s dissipation. During this period, 2033 counties experienced at least one TC-related event, impacting not only the eastern seaboard but also inland areas in central and southwestern regions. These findings underscore the importance of a multi-hazard perspective on TCs, enhancing hazard awareness and informing decision-making.
Each year, tropical cyclones (TCs) cause significant damage to both society and the environment through multiple hazards. While extensive research has focused on individual TC primary hazards, there is a gap in the comprehensive assessment of multiple TC-related hazards using a large number of TCs, likely due to limitations in methodology and data when identifying complex hazard interrelationships. To address this challenge, we developed novel methods using a rule-based natural language processing (NLP) approach to extract TC-related weather-hazardous events from official hazard records, of which narratives provide detailed characterization of TC multi-hazard interrelationships. We applied this method on TCs originating from the North Atlantic and East Pacific Oceans affecting the Contiguous United States. The results showed that the NLP methods, especially the large language models, can identify TC names from the texts with reasonable precision. A total of 21 488 events related to 179 TCs were identified from 2007 to 2022. These events encompassed 31 types of hazards, with precipitation and flooding being the most frequent and causing the most direct fatalities, while wind hazards cause the most direct property damage. Although TC wind and storm surges are typically concentrated within 100–200 km of the storm track, hazards can occur hundreds of kilometers away and even after the storm’s dissipation. During this period, 2033 counties experienced at least one TC-related event, impacting not only the eastern seaboard but also inland areas in central and southwestern regions. These findings underscore the importance of a multi-hazard perspective on TCs, enhancing hazard awareness and informing decision-making.
Assessing multi-hazards related to tropical cyclones through large language models and geospatial approaches
2024
Aufsatz (Zeitschrift)
Elektronische Ressource
Unbekannt
Metadata by DOAJ is licensed under CC BY-SA 1.0
Special issue on coastal hazards and risks due to tropical cyclones
| Taylor & Francis Verlag | 2022
Tropical cyclones activity analysis system
| British Library Conference Proceedings | 2000
QuikSCAT Wind Retrievals for Tropical Cyclones
| Online Contents | 2003