The Pearl River Delta represents a tide‐influenced lowland delta where sand mining has deepened the distributary channels. This study sets out to investigate the impacts of channel deepening on peak water levels across the delta. Based on collected hydrological data before and after sediment mining and hydrodynamic modeling, three regions are identified where sand mining‐induced changes in peak water levels have a different underlying cause. In the river‐dominated regions, peak water levels have significantly decreased because of the sand excavation, and thus the associated flood risk was reduced. In a transition region, the sand mining practice has increased peak water levels. In the tide‐dominated coastal region, the changes in peak water levels follow the rise in sea level and respond to human activities such as land reclamation. To understand the underlying mechanisms, the changes of subtidal friction governing tidal dynamics before and during sand mining are explored, and a simple linear regression model is used to predict subtidal water levels governing peak flood levels. Results from the hydrodynamic model and the linear regression models quantify exactly how the direct effect of channel bed lowering, river‐tide interaction, tidal amplification, and sea level rise have contributed to peak water level change, which offers a new analysis framework that can be applied in other systems.

Storm surges are among the deadliest natural hazards, but understanding and prediction of year-to-year variability of storm surges is challenging. Here, we demonstrate that the interannual variability of observed storm surge levels can be explained and further predicted, through a process-based study in Hong Kong. We find that El Niño-Southern Oscillation (ENSO) exerts a compound impact on storm surge levels through modulating tropical cyclones (TCs) and other forcing factors. The occurrence frequencies of local and remote TCs are responsible for the remaining variability in storm surge levels after removing the ENSO effect. Finally, we show that a statistical prediction model formed by ENSO and TC indices has good skill for prediction of extreme storm surge levels. The analysis approach can be applied to other coastal regions where tropical storms and the climate variability are main contributors to storm surges. Our study gives new insight into identifying “windows of opportunity” for successful prediction of storm surges on long-range timescales.