Abstract: The insufficient reliability of conventional monitoring methods under extreme rainfall conditions has resulted in an urgent need for new monitoring methods This study proposes an urban flood identification method that uses crawler technology (BeautifulSoup4) to extract flood-related terms from social media big data. The geographic information of water accumulation points was identified by combining BERT (Bidirectional Encoder Representations from Transformers) and CRF (Conditional Random Field) named entity recognition technology. Considering the extreme rainstorm and flood disaster in Zhengzhou on July 20, 2021 as an example, the accuracy of the identification method was verified by comparing it with the spatial distribution of the investigated water accumulation points and the numerical simulation of urban flooding. Results showed that the spatial distribution of the water accumulation points identified based on social media big data was basically consistent with that of the investigated water accumulation points, with a spatial overlap rate of 89.4%. Among the identified water accumulation points with spatial overlap, the proportions of water depths(H) proportions of 30＜ H ＜50, 50＜ H ＜100, 100＜ H ＜200, 200＜ H ＜300and H≥ 300 cm were 49.6%, 36.8%, 8.2%, 4.1% and 1.3%, respectively. The simulation results of the 50-year and 100-year design rainfall scenarios, as well as the measured rainfall scenarios on August 1, 2019, and July 22, 2024, demonstrate the rationality of the identification results in terms of spatial trends. Urban flood monitoring inversion based on social media big data is an important supplement to flood monitoring under extreme conditions and can be used as a crucial data support for validating numerical simulations on urban floods and emergency decision-making during flood disasters