To address the challenge of forecasting high tide levels in tidal river sections with frequent bedload transport, where stable bathymetric data are often lacking, this study developed an automatic forecasting model based on a “flood-tide separation-coupling” framework using only conventional hydrological data. The total water level of a station was first decomposed into “flood-controlled level” dominated by upstream runoff and “tide-controlled level” dominated by downstream tidal action, based on the bidirectional wave superposition principle. Four sub-models were then constructed using data of Zhijiang and Zhapu stations in Qiantang River from 2015 to 2019, including sub-models for the tide-controlled low tide level, the tide-controlled rising tide range, the flood-controlled level, and an empirical model for the effective tidal range propagation coefficient during floods.  The core of the model was to screen the "pure tide" and "pure runoff" samples through the flow threshold to calibrate the fundamental relationship. And a real-time calibration mechanism was designed, including a sliding time window (30 days) to dynamically update the tidal relationship and an offset coefficient (β) to adjust between baseline and envelope curves, enabling the model to adapt to riverbed scour and silting. Validation using data from 2020 to 2024 showed that during flood periods with a daily average flow greater than 10000 m³/s, the forecast error for high tide level of Zhijiang station was less than 0.3 m, with a maximum error of -0.28 m and a 100% qualification rate, meeting the Class A accuracy standard of the code for hydrological information prediction (GB/T 22482-2008). The model successfully eliminates the dependence on underwater topography, enhances forecasting stability in volatile riverbeds through real-time learning, and provides a practical, operational solution for floodwater level forecasting in similar tidal reaches.