Rainfall thresholds model for predicting urban local pluvial flooding

Study region

Guangzhou, China

Study focus

This study adapts the empirical ID/ED (Intensity-Duration/Accumulated Rainfall-Duration) threshold framework—commonly used for landslide prediction—to urban pluvial flood forecasting.

New hydrological insights

Traditional ID/ED thresholds face challenges in flood prediction due to hydrological complexity. This study evaluates their applicability to urban pluvial flooding, noting that existing empirical flood thresholds essentially remain within parameter optimization under the ID/ED framework rather than representing a theoretical change—supporting the framework’s applicability. To refine the model, the study also tested the inclusion of maximum rainfall intensity (I_max) and antecedent rainfall (E_a, reflecting pre-rainfall drainage conditions). However, stepwise regression analysis of 38–91 observed events from five stations rejected these variables. The proposed model uses accumulated event rainfall (E) and duration (D) to estimate flood peak depths (P), achieving a balance of accuracy (observation-based: R² = 0.66–0.89, RMSE = 0.028–0.104 m; simulation-based: R² = 0.89–0.99, RMSE = 0.005–0.072 m), generalizability, and robustness, while effectively minimizing overfitting. K-fold cross-validation ensured model stability, while classification modeling offered potential for performance improvement. The simple ED model structure improves flood risk communication for non-experts, balancing interpretability and feasibility. Though slightly less precise than conventional models, its operational advantages support disaster response in resource-limited areas, making it suitable for wider community-level use.