Unleashing the Power of Social Media: Advancing Weather Event Detection with VGI

By Dr. Sampath LonkaReviewed by Susha Cheriyedath, M.Sc.Jul 16 2023

The increasing frequency and intensity of extreme weather events, such as flooding, pose significant challenges. Access to accurate weather warnings and information about their impacts is crucial. Citizen volunteered geographic information (VGI) has emerged as a valuable complementary data source.

In a paper published in the journal Atmosphere, researchers proposed a visual analytics (VA) pipeline utilizing VGI from social media to enhance Sweden's impact-based weather warning system.

Background

With the rising occurrence of extreme weather events due to climate change, enhancing societal resilience and understanding local impacts have become crucial. Local and regional authorities worldwide prioritize predicting extreme weather, assessing its urban influences, and implementing adaptive actions. While official sensor networks provide valuable data, VGI has emerged as a complementary source. VGI encompasses participatory and opportunistic approaches, with social media data gaining importance. Leveraging VGI can provide localized information to improve disaster management and enhance impact-based weather warning systems.

Related work

The researchers reviewed previous works related to their development of the VA pipeline for detecting extreme weather events using text and images from VGI. Early visual analytics systems utilized messages generated by users on microblogging platforms to investigate and visualize crisis events. These systems predominantly use keyword-based Natural Language Processing (NLP) techniques to identify relevant tweets and emphasize visual exploration.

Later approaches integrated machine learning and data mining methods. For example, Chae et al. propose a VA emergency management technique incorporating topic modeling to track and extract topics from texts. Using Twitter data, Cerutti et al. use exploratory visualization and data mining to identify disaster-affected locations. Bosch et al. present ScatterBlogs2, a VA system utilizing filters, support vector machine classifiers, and topic modeling to extract and monitor relevant topics.

While these approaches have significantly contributed to crisis event detection by analyzing and visualizing microblog text, the researchers aim to advance the field by combining NLP, computer vision techniques, and interactive visualization in their VA pipeline. Their goal is to comprehensively explore various characteristics of extreme weather events sourced from Twitter's text and images.

Motivation

The ongoing research project AI4ClimateAdaptation describes the design considerations for the VA pipeline. The project aims to evaluate the incorporation of AI-based image and text analysis, VGI from citizens, and visualization to enhance weather warning processes and local impact knowledge.

The project aligns with Sweden's impact-based weather warning system, which involves direct consultation with regional and local stakeholders. The inclusion of local and regional actors enables more precise evaluations of local thresholds and risk factors, as well as the creation of assessment support documentation. VGI, including participatory and opportunistic forms, is of great interest in this context.

VA pipeline design

The design of the VA pipeline for identifying and exploring extreme weather events from VGI involves several key considerations. The pipeline encompasses multiple facets, each with its own set of challenges and opportunities.

Data collection: This is a crucial aspect of the pipeline design. It involves sourcing data from various channels, including public weather warning announcements and social media platforms like Twitter. Custom data collection methods are explored, such as utilizing the Twitter Streaming API and developing a citizen sensing application for image submissions. The goal is to identify relevant subsets of data and establish effective collection approaches.

Classification: Another critical step focuses on identifying flooding events through the analysis of flood-related VGI texts and images. The researchers utilized training data from publicly available labeled datasets and case-specific data collection efforts. They considered transfer learning and a range of machine learning approaches, including traditional methods and deep learning models, for accurate classification.

Text classification involves categorizing VGI text entries as flood-relevant or flood-irrelevant and performing further analysis. The researchers employed multi-label classification models and progressive analysis techniques to extract relevant datasets and perform additional tasks such as topic modeling and sentiment analysis. The choice of classification methods depends on data modality, availability of training data, and desired categorization outcomes.

Location extraction: Location extraction plays a vital role in connecting VGI to geographical locations. Geoparsing techniques are used to extract toponyms from text and associate them with real-world coordinates, addressing challenges like toponym ambiguity and limited geographic attributes in social media posts. Geotagging methods are employed to retrieve location information from images and videos.

Interactive visual interface: The final step involves designing an interactive visual interface that integrates the components of the VA pipeline. Considerations include the data, users, and analysis tasks. Stakeholder engagement and co-design processes help assess user needs and define requirements. The visual interface includes map views, content views, temporal views, and potentially impact-related information views to facilitate tasks such as identifying weather event occurrences, extents, and impacts.

Conclusion

In conclusion, the researchers of this study presented the design considerations and opportunities for a VA pipeline to identify and explore extreme weather events, particularly floods, using VGI from social media. The work aims to support stakeholders involved in the regional consultation process for weather warnings. The outlined design space, considerations, and limitations serve as the foundation for future work, including the implementation and assessment of the VA pipeline with representative experts.