Multi-data source based quantifying urban flood severity in major Chinese cities (2000–2024) using a hybrid machine-learning weighting framework

Abstract. Urban flooding poses a major challenge to sustainable urban development, yet most existing assessments focus on single cities or river basins and rely on limited historical records. This study integrates multi-source data from 20 Chinese cities over 2000–2024 to develop a comparable long-term assessment of urban flood severity. To address the fragmentation and inconsistency of flood evidence across official records, news reports, and social media, we construct an event-level database and derive a Flood Severity Index (FSI) using an interpretable data-driven weighting and ensemble framework. Robustness is evaluated through repeated resampling and consistency checks across cities and years. The results show that southern cities experience more frequent and severe flooding, whereas northern cities are generally less affected but more vulnerable to abrupt extremes. These findings suggest distinct governance priorities: reducing chronic exposure in southern cities and strengthening preparedness for high-impact shocks in northern cities. The proposed framework is transferable to other regions and provides a basis for future cross-regional flood risk comparison and adaptive urban risk governance.