Deep learning-based assessment of flood severity using social media streams

With a rapid change in climate, flood events have become a common issue in many parts of India. Cities like Pune, Chennai experience heavy rainfall every year, followed by devastating floods. To better support the flood emergency plan operations, it is essential to have real-time flood maps depicting flood levels across the city. It can be made possible by mining information from flood-related social media posts shared by the public during floods. In this paper, we propose a deep learning-based method to assess the flood severity by using text and image data extracted from the social media posts. In the first stage of the methodology, the text data from the social media posts are analyzed, and flood-related posts are then passed to the second stage. In the second stage, the images associated with the social media posts are analyzed, and the severity of the flood in the particular location is updated in real-time. The text and image classification models are trained using the social media feeds posted during Pune, Chennai, and Kerala floods. The accuracy obtained using the proposed methodology is 98% and 78% for text and image classification, respectively. By introducing text classification in the flood severity estimation task, social media posts that are irrelevant to the flood severity estimation task are ignored without processing the multimedia data associated with it. This, in turn, results in reduced usage of valuable computational resources as classification of multimedia data is expensive compared to the classification of microblog text data.