Traditional methods for assessing urban waterlogging hazards lack real-time efficiency. This study develops a rapid hazard assessment method for urban waterlogging events using social media data and large language models, with the urban waterlogging event in Guilin in 2024 as a case study. We constructed a keyword database for urban waterlogging, combined with web crawling techniques to collect related data from social media platforms. This study applied the BERT-BILSTM-CRF model, the GPT-3.5-turbo model, and the ERNIE-3.5-128K model to extract waterlogging locations, comparing them with the official waterlogging location data to validate and select the optimal model. The model was then used for urban waterlogging hazard assessment at the township scale. The results indicate that the establishment of a keyword database facilitates access to more social media data. The GPT-3.5-turbo model demonstrates superior performance in waterlogging location identification, with a coverage of 93% compared with the actual waterlogging locations, and an F1 score of 0.8066. During the Guilin 6.19 waterlogging, areas with high hazard intensities mainly concentrated in the city center; meanwhile, the hazard intensities of some townships in the suburbs intensified from 17 June to 21 June. This study validates the effectiveness and timeliness of using social media data and large language models for rapid hazard assessment of urban waterlogging events through practical cases, which can provide reliable support for future urban disaster prevention and reduction efforts.

With the acceleration of global climate change and urbanization, urban resilience has become a critical issue. This study, based on the Pressure-State-Response (PSR) model, constructs an urban resilience evaluation index system for Sanming City. Indicator weights are determined by combining the Analytic Hierarchy Process (AHP) and the entropy weight method. Spatial analysis methods, such as spatial autocorrelation, kernel density estimation, standard deviation ellipses, and geographic detectors, are employed to explore spatial–temporal analysis and driving factors of urban resilience. The results show the following: (1) from 2014 to 2022, Sanming’s urban resilience index initially increased and then declined; (2) the spatial distribution of urban resilience is uneven, with high-resilience areas concentrated in the city center and southeast, while the northwest is relatively low; (3) Local Moran’s I analysis confirms significant positive spatial autocorrelation, with regional differences gradually expanding; (4) geographic detector analysis reveals that NDVI, monthly maximum precipitation, nighttime light index, annual average PM2.5 concentration, and impervious surface ratio are key drivers of urban resilience; (5) factor interactions show nonlinear enhancement, with ecological and climatic–environmental factors interacting as key drivers of urban resilience changes.

Flood disasters exert profound global impacts on socioeconomic systems and ecosystems, creating an urgent need for high-quality datasets to support flood detection. However, existing datasets remain constrained by limited source diversity, annotation precision, and scale. Sen2GF3Floods is introduced as the first dataset that integrates pre-disaster Sentinel-2 optical imagery with post-disaster Gaofen-3 SAR imagery. A dual-temporal collaborative annotation framework was constructed, combining semi-automatic labeling with active learning based on U-Net++ to achieve an effective balance between accuracy and efficiency. The dataset contains 21,483 standardized samples from nine major flood events, covering diverse geomorphological settings. Benchmark evaluations using five semantic segmentation models including U-Net, U-Net++, DeepLabV3+, DANet, and SegFormer indicate that multi-source fusion of Sentinel-2 RGB and NIR data with GF-3 HH and HV data delivers robust flood mapping performance across varied scenarios. Active learning reduces annotation costs while maintaining quality, and models trained on GF-3 SAR demonstrate good transferability to Sentinel-1 SAR. These results establish Sen2GF3Floods as a valuable resource for advancing flood detection algorithms and supporting operational and real-time disaster response.

Loess landslide disasters are influenced by a multitude of factors, including slope conditions, triggering mechanisms, and spatial attributes. Extracting these factors from unstructured geological texts is challenging due to nested entities, semantic ambiguity, and rare domain-specific terms. This study proposes a joint extraction framework guided by a domain ontology that categorizes six types of loess landslide influencing factors, including spatial relationships. The ontology facilitates conceptual classification and semi-automatic nested entity annotation, enabling the construction of a high-quality corpus with eight tag types. The model integrates a Soft-Lexicon mechanism that enhances character-level GloVe embeddings with explicit lexical features, including domain terms, part-of-speech tags, and word boundary indicators derived from a domain-specific lexicon. The resulting hybrid character-level representations are then fed into a BiLSTM-CRF architecture to jointly extract entities, attributes, and multi-level spatial and causal relationships. Extracted results are structured using a content-knowledge model to build a spatially enriched knowledge graph, supporting semantic queries and intelligent reasoning. Experimental results demonstrate improved performance over baseline methods, showcasing the framework’s effectiveness in geohazard information extraction and disaster risk analysis.