A Bibliometric-Qualitative Literature Review of Flood Research Based on Deep Learning

Floods are a natural disaster that frequently occurs worldwide and has a high destructive power, with complex influencing factors. Establishing flood models for monitoring, simulation, estimation and prediction is crucial to reducing disaster risk and minimizing human and property losses. With the development of computer computing power and deep learning's powerful classification and prediction capabilities, the application of deep learning in flood research is increasingly attracting attention, leading to a significant increase in data-driven flood research using deep learning. This article uses a mixed method combining bibliometric analysis and qualitative analysis to analyze the characteristics and development trends of deep learning-based flood research. Using a bibliometric analysis of 425 papers in the Web of Science core database, scientific knowledge graph analysis was conducted, indicating that the field is entering a period of rapid development with a promising future. Based on the latest literature and combined with the bibliometric results,23 selected papers were qualitatively and systematically analyzed to accurately evaluate the current situation and future directions of deep learning in flood research. Three main research scenarios emerged from the qualitative results: flood sensitivity, flood inundation, and precipitation runoff. Combining recent research trends, the data sources, method choices and strategies for different research scenarios were summarized, and suggestions for future research development and innovation were proposed from three levels of data, algorithms, and research contexts to promote the further development and application of deep learning in flood research.