Scholarlink

<p>Effective flood management in urban planning relies on accurate, timely data, which can be sourced from social media platforms for real-time post-flood damage information. However, many social media content lack location, creating a significant challenge for spatial analysis. This study addresses this gap by proposing a novel framework to infer the locations of post-flood events extracted from social media content, leveraging flood vulnerability maps and a structured knowledge base. The methodology involves four key steps, extracting flood-related events using hypergraph-based clustering; creating bounding boxes for potential event locations by integrating flood-related keywords, spatial proximity, and temporal patterns; constructing a knowledge base incorporating flood vulnerability criteria; and inferring event locations by comparing non-geo-tagged events against the knowledge base rules and aligning them with spatiotemporal bounding boxes. By analyzing 150,000 social media posts from flood-affected regions in southwestern Iran, such as Ahvaz, between April 6–16, 2019, the method identified 27 flood and 1200 post-flood events; of the 970 non-geo-tagged events, 69 were inferred inside the study region, while the remaining 901 were inferred out-of-region and were not mapped. Evaluation metrics, including 70% Precision, 77% Recall, and 74% F1 Score, show the model’s effectiveness in flood event detection, while spatial accuracy metrics, such as 2.15&#xa0;km Mean Error Distance and 0.65 Mean Intersection over Union, confirm its reliability in location inference. The study highlights social media data’s potential for real-time flood management, especially in areas with scarce geotagged content.</p>

错误：搜索内容不能为空，请输入英文关键词
错误：关键词超出字数限制，请精简
高级检索

Leveraging social media and vulnerability maps for post-flood event localization

  • 期刊论文

  • 发表日期：

    2026年6月29日

查看全文
  • Hossein Bahadorizadeh,
  • Mohammad Reza Malek

摘要

Effective flood management in urban planning relies on accurate, timely data, which can be sourced from social media platforms for real-time post-flood damage information. However, many social media content lack location, creating a significant challenge for spatial analysis. This study addresses this gap by proposing a novel framework to infer the locations of post-flood events extracted from social media content, leveraging flood vulnerability maps and a structured knowledge base. The methodology involves four key steps, extracting flood-related events using hypergraph-based clustering; creating bounding boxes for potential event locations by integrating flood-related keywords, spatial proximity, and temporal patterns; constructing a knowledge base incorporating flood vulnerability criteria; and inferring event locations by comparing non-geo-tagged events against the knowledge base rules and aligning them with spatiotemporal bounding boxes. By analyzing 150,000 social media posts from flood-affected regions in southwestern Iran, such as Ahvaz, between April 6–16, 2019, the method identified 27 flood and 1200 post-flood events; of the 970 non-geo-tagged events, 69 were inferred inside the study region, while the remaining 901 were inferred out-of-region and were not mapped. Evaluation metrics, including 70% Precision, 77% Recall, and 74% F1 Score, show the model’s effectiveness in flood event detection, while spatial accuracy metrics, such as 2.15 km Mean Error Distance and 0.65 Mean Intersection over Union, confirm its reliability in location inference. The study highlights social media data’s potential for real-time flood management, especially in areas with scarce geotagged content.