Flood damage estimation and disaster perception using multi-source data: a case study of Henan, China

Abstract

Urban flood is becoming increasingly frequent worldwide, posing a serious threat to life and property. Accurately assessing flood damage is critical for risk management and flood prevention. However, traditional estimation methods are resource-intensive and untimely. Previous studies have not combined flood damage estimation with public sentiment attribution analysis into a framework, nor have they quantified the driving factors of sentiment changes. Therefore, this study developed a novel framework for estimating flood damage and public perception. First, based on historical flood data, the genetic algorithm (GA)-back propagation (BP) model was optimized using the price base-setting (PBS) method to estimate flood damage. Real-time flood information was then extracted from social media data using the GloVe model and Kruskal algorithm. The Naive Bayes algorithm was used for sentiment analysis, while the Ordinary Least Squares regression was used to analyze the factors influencing changes in public sentiment. The results suggest that the PBS-GA-BP model is more accurate in its estimation (MAE = 4.24 and RMSE = 5.45) than the GA-BP model (MAE = 20.45 and RMSE = 25.27). Sentiment is influenced by emergency rescue operations and property damage, particularly vehicle damage. People in flood-affected areas tend to express more negative sentiment. This study provides valuable insights for flood management and recovery efforts.