Estimating urban flood severity using social media text and remote sensing data: a case study of Henan, China

Rapid estimation of urban flood severity (UFS) can support decision-making in urban planning and emergency management. However, traditional methods face challenges in analyzing the social impact of UFS, such as the affected population. To overcome these challenges, this study proposed a hybrid framework that combines textual data with remote sensing data to estimate UFS. Taking the 7.20 flood in Henan as a case study, social media text was collected. A text classification system was constructed using three algorithms, including Naive Bayes. Second, the level of flood severity (LFS) model was used to estimate UFS. Third, using remote sensing data to identify urban flood-affected areas and estimate direct economic losses and affected population. Finally, verifying the accuracy of the results. The results show: (1) UFS in Zhengzhou and Xinxiang is extremely severe. The number of affected population is the largest, followed by property damage and infrastructure damage. (2) The MRE of direct economic losses is 9.67% and the RMSE is 102.27. The RE of affected population is 7.30%. (3) The results obtained from the hybrid framework are basically consistent, with a correlation coefficient of 0.86. This study contributes to the UFS estimation, flood response and risk management under climate change scenarios.