U-RNN high-resolution spatiotemporal nowcasting of urban flooding

Xiaoyan Cao, Baoying Wang, Yao Yao, Lin Zhang, Yanwen Xing, Junqi Mao, Runqiao Zhang, Guangtao Fu, Alistair G.L. Borthwick, Huapeng Qin

Abstract: Highlights•A novel deep learning (DL) model (U-RNN) is proposed for urban flood nowcasting.•A new DL training paradigm supports long-sequence generalization with low resources.•U-RNN excels in predicting flood extent, peak depth, hydrographs with high accuracy.•U-RNN nowcasts ahead 6 h at 2 m and 1 min resolution, 10× longer than baseline.•U-RNN is 100x faster than the numerical model and 50% more accurate than baselines.

External IDs:doi:10.1016/j.jhydrol.2025.133117